Skip to content

Detections

The sv.Detections class in the Supervision library standardizes results from various object detection and segmentation models into a consistent format. This class simplifies data manipulation and filtering, providing a uniform API for integration with Supervision trackers, annotators, and tools.

Use sv.Detections.from_inference method, which accepts model results from both detection and segmentation models.

import cv2
import supervision as sv
from inference import get_model

model = get_model(model_id="yolov8n-640")
image = cv2.imread(<SOURCE_IMAGE_PATH>)
results = model.infer(image)[0]
detections = sv.Detections.from_inference(results)

Use sv.Detections.from_ultralytics method, which accepts model results from both detection and segmentation models.

import cv2
import supervision as sv
from ultralytics import YOLO

model = YOLO("yolov8n.pt")
image = cv2.imread(<SOURCE_IMAGE_PATH>)
results = model(image)[0]
detections = sv.Detections.from_ultralytics(results)

Use sv.Detections.from_transformers method, which accepts model results from both detection and segmentation models.

import torch
import supervision as sv
from PIL import Image
from transformers import DetrImageProcessor, DetrForObjectDetection

processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")

image = Image.open(<SOURCE_IMAGE_PATH>)
inputs = processor(images=image, return_tensors="pt")

with torch.no_grad():
    outputs = model(**inputs)

width, height = image.size
target_size = torch.tensor([[height, width]])
results = processor.post_process_object_detection(
    outputs=outputs, target_sizes=target_size)[0]
detections = sv.Detections.from_transformers(
    transformers_results=results,
    id2label=model.config.id2label)

Attributes:

Name Type Description
xyxy ndarray

An array of shape (n, 4) containing the bounding boxes coordinates in format [x1, y1, x2, y2]

mask Optional[ndarray]

(Optional[np.ndarray]): An array of shape (n, H, W) containing the segmentation masks (bool data type).

confidence Optional[ndarray]

An array of shape (n,) containing the confidence scores of the detections.

class_id Optional[ndarray]

An array of shape (n,) containing the class ids of the detections.

tracker_id Optional[ndarray]

An array of shape (n,) containing the tracker ids of the detections.

data Dict[str, Union[ndarray, List]]

A dictionary containing additional data where each key is a string representing the data type, and the value is either a NumPy array or a list of corresponding data.

metadata Dict[str, Any]

A dictionary containing collection-level metadata that applies to the entire set of detections. This may include information such as the video name, camera parameters, timestamp, or other global metadata.

Source code in supervision/detection/core.py
  46
  47
  48
  49
  50
  51
  52
  53
  54
  55
  56
  57
  58
  59
  60
  61
  62
  63
  64
  65
  66
  67
  68
  69
  70
  71
  72
  73
  74
  75
  76
  77
  78
  79
  80
  81
  82
  83
  84
  85
  86
  87
  88
  89
  90
  91
  92
  93
  94
  95
  96
  97
  98
  99
 100
 101
 102
 103
 104
 105
 106
 107
 108
 109
 110
 111
 112
 113
 114
 115
 116
 117
 118
 119
 120
 121
 122
 123
 124
 125
 126
 127
 128
 129
 130
 131
 132
 133
 134
 135
 136
 137
 138
 139
 140
 141
 142
 143
 144
 145
 146
 147
 148
 149
 150
 151
 152
 153
 154
 155
 156
 157
 158
 159
 160
 161
 162
 163
 164
 165
 166
 167
 168
 169
 170
 171
 172
 173
 174
 175
 176
 177
 178
 179
 180
 181
 182
 183
 184
 185
 186
 187
 188
 189
 190
 191
 192
 193
 194
 195
 196
 197
 198
 199
 200
 201
 202
 203
 204
 205
 206
 207
 208
 209
 210
 211
 212
 213
 214
 215
 216
 217
 218
 219
 220
 221
 222
 223
 224
 225
 226
 227
 228
 229
 230
 231
 232
 233
 234
 235
 236
 237
 238
 239
 240
 241
 242
 243
 244
 245
 246
 247
 248
 249
 250
 251
 252
 253
 254
 255
 256
 257
 258
 259
 260
 261
 262
 263
 264
 265
 266
 267
 268
 269
 270
 271
 272
 273
 274
 275
 276
 277
 278
 279
 280
 281
 282
 283
 284
 285
 286
 287
 288
 289
 290
 291
 292
 293
 294
 295
 296
 297
 298
 299
 300
 301
 302
 303
 304
 305
 306
 307
 308
 309
 310
 311
 312
 313
 314
 315
 316
 317
 318
 319
 320
 321
 322
 323
 324
 325
 326
 327
 328
 329
 330
 331
 332
 333
 334
 335
 336
 337
 338
 339
 340
 341
 342
 343
 344
 345
 346
 347
 348
 349
 350
 351
 352
 353
 354
 355
 356
 357
 358
 359
 360
 361
 362
 363
 364
 365
 366
 367
 368
 369
 370
 371
 372
 373
 374
 375
 376
 377
 378
 379
 380
 381
 382
 383
 384
 385
 386
 387
 388
 389
 390
 391
 392
 393
 394
 395
 396
 397
 398
 399
 400
 401
 402
 403
 404
 405
 406
 407
 408
 409
 410
 411
 412
 413
 414
 415
 416
 417
 418
 419
 420
 421
 422
 423
 424
 425
 426
 427
 428
 429
 430
 431
 432
 433
 434
 435
 436
 437
 438
 439
 440
 441
 442
 443
 444
 445
 446
 447
 448
 449
 450
 451
 452
 453
 454
 455
 456
 457
 458
 459
 460
 461
 462
 463
 464
 465
 466
 467
 468
 469
 470
 471
 472
 473
 474
 475
 476
 477
 478
 479
 480
 481
 482
 483
 484
 485
 486
 487
 488
 489
 490
 491
 492
 493
 494
 495
 496
 497
 498
 499
 500
 501
 502
 503
 504
 505
 506
 507
 508
 509
 510
 511
 512
 513
 514
 515
 516
 517
 518
 519
 520
 521
 522
 523
 524
 525
 526
 527
 528
 529
 530
 531
 532
 533
 534
 535
 536
 537
 538
 539
 540
 541
 542
 543
 544
 545
 546
 547
 548
 549
 550
 551
 552
 553
 554
 555
 556
 557
 558
 559
 560
 561
 562
 563
 564
 565
 566
 567
 568
 569
 570
 571
 572
 573
 574
 575
 576
 577
 578
 579
 580
 581
 582
 583
 584
 585
 586
 587
 588
 589
 590
 591
 592
 593
 594
 595
 596
 597
 598
 599
 600
 601
 602
 603
 604
 605
 606
 607
 608
 609
 610
 611
 612
 613
 614
 615
 616
 617
 618
 619
 620
 621
 622
 623
 624
 625
 626
 627
 628
 629
 630
 631
 632
 633
 634
 635
 636
 637
 638
 639
 640
 641
 642
 643
 644
 645
 646
 647
 648
 649
 650
 651
 652
 653
 654
 655
 656
 657
 658
 659
 660
 661
 662
 663
 664
 665
 666
 667
 668
 669
 670
 671
 672
 673
 674
 675
 676
 677
 678
 679
 680
 681
 682
 683
 684
 685
 686
 687
 688
 689
 690
 691
 692
 693
 694
 695
 696
 697
 698
 699
 700
 701
 702
 703
 704
 705
 706
 707
 708
 709
 710
 711
 712
 713
 714
 715
 716
 717
 718
 719
 720
 721
 722
 723
 724
 725
 726
 727
 728
 729
 730
 731
 732
 733
 734
 735
 736
 737
 738
 739
 740
 741
 742
 743
 744
 745
 746
 747
 748
 749
 750
 751
 752
 753
 754
 755
 756
 757
 758
 759
 760
 761
 762
 763
 764
 765
 766
 767
 768
 769
 770
 771
 772
 773
 774
 775
 776
 777
 778
 779
 780
 781
 782
 783
 784
 785
 786
 787
 788
 789
 790
 791
 792
 793
 794
 795
 796
 797
 798
 799
 800
 801
 802
 803
 804
 805
 806
 807
 808
 809
 810
 811
 812
 813
 814
 815
 816
 817
 818
 819
 820
 821
 822
 823
 824
 825
 826
 827
 828
 829
 830
 831
 832
 833
 834
 835
 836
 837
 838
 839
 840
 841
 842
 843
 844
 845
 846
 847
 848
 849
 850
 851
 852
 853
 854
 855
 856
 857
 858
 859
 860
 861
 862
 863
 864
 865
 866
 867
 868
 869
 870
 871
 872
 873
 874
 875
 876
 877
 878
 879
 880
 881
 882
 883
 884
 885
 886
 887
 888
 889
 890
 891
 892
 893
 894
 895
 896
 897
 898
 899
 900
 901
 902
 903
 904
 905
 906
 907
 908
 909
 910
 911
 912
 913
 914
 915
 916
 917
 918
 919
 920
 921
 922
 923
 924
 925
 926
 927
 928
 929
 930
 931
 932
 933
 934
 935
 936
 937
 938
 939
 940
 941
 942
 943
 944
 945
 946
 947
 948
 949
 950
 951
 952
 953
 954
 955
 956
 957
 958
 959
 960
 961
 962
 963
 964
 965
 966
 967
 968
 969
 970
 971
 972
 973
 974
 975
 976
 977
 978
 979
 980
 981
 982
 983
 984
 985
 986
 987
 988
 989
 990
 991
 992
 993
 994
 995
 996
 997
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
@dataclass
class Detections:
    """
    The `sv.Detections` class in the Supervision library standardizes results from
    various object detection and segmentation models into a consistent format. This
    class simplifies data manipulation and filtering, providing a uniform API for
    integration with Supervision [trackers](/trackers/), [annotators](/latest/detection/annotators/), and [tools](/detection/tools/line_zone/).

    === "Inference"

        Use [`sv.Detections.from_inference`](/detection/core/#supervision.detection.core.Detections.from_inference)
        method, which accepts model results from both detection and segmentation models.

        ```python
        import cv2
        import supervision as sv
        from inference import get_model

        model = get_model(model_id="yolov8n-640")
        image = cv2.imread(<SOURCE_IMAGE_PATH>)
        results = model.infer(image)[0]
        detections = sv.Detections.from_inference(results)
        ```

    === "Ultralytics"

        Use [`sv.Detections.from_ultralytics`](/detection/core/#supervision.detection.core.Detections.from_ultralytics)
        method, which accepts model results from both detection and segmentation models.

        ```python
        import cv2
        import supervision as sv
        from ultralytics import YOLO

        model = YOLO("yolov8n.pt")
        image = cv2.imread(<SOURCE_IMAGE_PATH>)
        results = model(image)[0]
        detections = sv.Detections.from_ultralytics(results)
        ```

    === "Transformers"

        Use [`sv.Detections.from_transformers`](/detection/core/#supervision.detection.core.Detections.from_transformers)
        method, which accepts model results from both detection and segmentation models.

        ```python
        import torch
        import supervision as sv
        from PIL import Image
        from transformers import DetrImageProcessor, DetrForObjectDetection

        processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
        model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")

        image = Image.open(<SOURCE_IMAGE_PATH>)
        inputs = processor(images=image, return_tensors="pt")

        with torch.no_grad():
            outputs = model(**inputs)

        width, height = image.size
        target_size = torch.tensor([[height, width]])
        results = processor.post_process_object_detection(
            outputs=outputs, target_sizes=target_size)[0]
        detections = sv.Detections.from_transformers(
            transformers_results=results,
            id2label=model.config.id2label)
        ```

    Attributes:
        xyxy (np.ndarray): An array of shape `(n, 4)` containing
            the bounding boxes coordinates in format `[x1, y1, x2, y2]`
        mask: (Optional[np.ndarray]): An array of shape
            `(n, H, W)` containing the segmentation masks (`bool` data type).
        confidence (Optional[np.ndarray]): An array of shape
            `(n,)` containing the confidence scores of the detections.
        class_id (Optional[np.ndarray]): An array of shape
            `(n,)` containing the class ids of the detections.
        tracker_id (Optional[np.ndarray]): An array of shape
            `(n,)` containing the tracker ids of the detections.
        data (Dict[str, Union[np.ndarray, List]]): A dictionary containing additional
            data where each key is a string representing the data type, and the value
            is either a NumPy array or a list of corresponding data.
        metadata (Dict[str, Any]): A dictionary containing collection-level metadata
            that applies to the entire set of detections. This may include information such
            as the video name, camera parameters, timestamp, or other global metadata.
    """  # noqa: E501 // docs

    xyxy: np.ndarray
    mask: Optional[np.ndarray] = None
    confidence: Optional[np.ndarray] = None
    class_id: Optional[np.ndarray] = None
    tracker_id: Optional[np.ndarray] = None
    data: Dict[str, Union[np.ndarray, List]] = field(default_factory=dict)
    metadata: Dict[str, Any] = field(default_factory=dict)

    def __post_init__(self):
        validate_detections_fields(
            xyxy=self.xyxy,
            mask=self.mask,
            confidence=self.confidence,
            class_id=self.class_id,
            tracker_id=self.tracker_id,
            data=self.data,
        )

    def __len__(self):
        """
        Returns the number of detections in the Detections object.
        """
        return len(self.xyxy)

    def __iter__(
        self,
    ) -> Iterator[
        Tuple[
            np.ndarray,
            Optional[np.ndarray],
            Optional[float],
            Optional[int],
            Optional[int],
            Dict[str, Union[np.ndarray, List]],
        ]
    ]:
        """
        Iterates over the Detections object and yield a tuple of
        `(xyxy, mask, confidence, class_id, tracker_id, data)` for each detection.
        """
        for i in range(len(self.xyxy)):
            yield (
                self.xyxy[i],
                self.mask[i] if self.mask is not None else None,
                self.confidence[i] if self.confidence is not None else None,
                self.class_id[i] if self.class_id is not None else None,
                self.tracker_id[i] if self.tracker_id is not None else None,
                get_data_item(self.data, i),
            )

    def __eq__(self, other: Detections):
        return all(
            [
                np.array_equal(self.xyxy, other.xyxy),
                np.array_equal(self.mask, other.mask),
                np.array_equal(self.class_id, other.class_id),
                np.array_equal(self.confidence, other.confidence),
                np.array_equal(self.tracker_id, other.tracker_id),
                is_data_equal(self.data, other.data),
                is_metadata_equal(self.metadata, other.metadata),
            ]
        )

    @classmethod
    def from_yolov5(cls, yolov5_results) -> Detections:
        """
        Creates a Detections instance from a
        [YOLOv5](https://github.com/ultralytics/yolov5) inference result.

        Args:
            yolov5_results (yolov5.models.common.Detections):
                The output Detections instance from YOLOv5

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import cv2
            import torch
            import supervision as sv

            image = cv2.imread(<SOURCE_IMAGE_PATH>)
            model = torch.hub.load('ultralytics/yolov5', 'yolov5s')
            result = model(image)
            detections = sv.Detections.from_yolov5(result)
            ```
        """
        yolov5_detections_predictions = yolov5_results.pred[0].cpu().cpu().numpy()

        return cls(
            xyxy=yolov5_detections_predictions[:, :4],
            confidence=yolov5_detections_predictions[:, 4],
            class_id=yolov5_detections_predictions[:, 5].astype(int),
        )

    @classmethod
    def from_ultralytics(cls, ultralytics_results) -> Detections:
        """
        Creates a `sv.Detections` instance from a
        [YOLOv8](https://github.com/ultralytics/ultralytics) inference result.

        !!! Note

            `from_ultralytics` is compatible with
            [detection](https://docs.ultralytics.com/tasks/detect/),
            [segmentation](https://docs.ultralytics.com/tasks/segment/), and
            [OBB](https://docs.ultralytics.com/tasks/obb/) models.

        Args:
            ultralytics_results (ultralytics.yolo.engine.results.Results):
                The output Results instance from Ultralytics

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import cv2
            import supervision as sv
            from ultralytics import YOLO

            image = cv2.imread(<SOURCE_IMAGE_PATH>)
            model = YOLO('yolov8s.pt')
            results = model(image)[0]
            detections = sv.Detections.from_ultralytics(results)
            ```
        """

        if hasattr(ultralytics_results, "obb") and ultralytics_results.obb is not None:
            class_id = ultralytics_results.obb.cls.cpu().numpy().astype(int)
            class_names = np.array([ultralytics_results.names[i] for i in class_id])
            oriented_box_coordinates = ultralytics_results.obb.xyxyxyxy.cpu().numpy()
            return cls(
                xyxy=ultralytics_results.obb.xyxy.cpu().numpy(),
                confidence=ultralytics_results.obb.conf.cpu().numpy(),
                class_id=class_id,
                tracker_id=ultralytics_results.obb.id.int().cpu().numpy()
                if ultralytics_results.obb.id is not None
                else None,
                data={
                    ORIENTED_BOX_COORDINATES: oriented_box_coordinates,
                    CLASS_NAME_DATA_FIELD: class_names,
                },
            )

        if hasattr(ultralytics_results, "boxes") and ultralytics_results.boxes is None:
            masks = extract_ultralytics_masks(ultralytics_results)
            return cls(
                xyxy=mask_to_xyxy(masks),
                mask=masks,
                class_id=np.arange(len(ultralytics_results)),
            )

        class_id = ultralytics_results.boxes.cls.cpu().numpy().astype(int)
        class_names = np.array([ultralytics_results.names[i] for i in class_id])
        return cls(
            xyxy=ultralytics_results.boxes.xyxy.cpu().numpy(),
            confidence=ultralytics_results.boxes.conf.cpu().numpy(),
            class_id=class_id,
            mask=extract_ultralytics_masks(ultralytics_results),
            tracker_id=ultralytics_results.boxes.id.int().cpu().numpy()
            if ultralytics_results.boxes.id is not None
            else None,
            data={CLASS_NAME_DATA_FIELD: class_names},
        )

    @classmethod
    def from_yolo_nas(cls, yolo_nas_results) -> Detections:
        """
        Creates a Detections instance from a
        [YOLO-NAS](https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS.md)
        inference result.

        Args:
            yolo_nas_results (ImageDetectionPrediction):
                The output Results instance from YOLO-NAS
                ImageDetectionPrediction is coming from
                'super_gradients.training.models.prediction_results'

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import cv2
            from super_gradients.training import models
            import supervision as sv

            image = cv2.imread(<SOURCE_IMAGE_PATH>)
            model = models.get('yolo_nas_l', pretrained_weights="coco")

            result = list(model.predict(image, conf=0.35))[0]
            detections = sv.Detections.from_yolo_nas(result)
            ```
        """
        if np.asarray(yolo_nas_results.prediction.bboxes_xyxy).shape[0] == 0:
            return cls.empty()

        return cls(
            xyxy=yolo_nas_results.prediction.bboxes_xyxy,
            confidence=yolo_nas_results.prediction.confidence,
            class_id=yolo_nas_results.prediction.labels.astype(int),
        )

    @classmethod
    def from_tensorflow(
        cls, tensorflow_results: dict, resolution_wh: tuple
    ) -> Detections:
        """
        Creates a Detections instance from a
        [Tensorflow Hub](https://www.tensorflow.org/hub/tutorials/tf2_object_detection)
        inference result.

        Args:
            tensorflow_results (dict):
                The output results from Tensorflow Hub.

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import tensorflow as tf
            import tensorflow_hub as hub
            import numpy as np
            import cv2

            module_handle = "https://tfhub.dev/tensorflow/centernet/hourglass_512x512_kpts/1"
            model = hub.load(module_handle)
            img = np.array(cv2.imread(SOURCE_IMAGE_PATH))
            result = model(img)
            detections = sv.Detections.from_tensorflow(result)
            ```
        """

        boxes = tensorflow_results["detection_boxes"][0].numpy()
        boxes[:, [0, 2]] *= resolution_wh[0]
        boxes[:, [1, 3]] *= resolution_wh[1]
        boxes = boxes[:, [1, 0, 3, 2]]
        return cls(
            xyxy=boxes,
            confidence=tensorflow_results["detection_scores"][0].numpy(),
            class_id=tensorflow_results["detection_classes"][0].numpy().astype(int),
        )

    @classmethod
    def from_deepsparse(cls, deepsparse_results) -> Detections:
        """
        Creates a Detections instance from a
        [DeepSparse](https://github.com/neuralmagic/deepsparse)
        inference result.

        Args:
            deepsparse_results (deepsparse.yolo.schemas.YOLOOutput):
                The output Results instance from DeepSparse.

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import supervision as sv
            from deepsparse import Pipeline

            yolo_pipeline = Pipeline.create(
                task="yolo",
                model_path = "zoo:cv/detection/yolov5-l/pytorch/ultralytics/coco/pruned80_quant-none"
             )
            result = yolo_pipeline(<SOURCE IMAGE PATH>)
            detections = sv.Detections.from_deepsparse(result)
            ```
        """  # noqa: E501 // docs

        if np.asarray(deepsparse_results.boxes[0]).shape[0] == 0:
            return cls.empty()

        return cls(
            xyxy=np.array(deepsparse_results.boxes[0]),
            confidence=np.array(deepsparse_results.scores[0]),
            class_id=np.array(deepsparse_results.labels[0]).astype(float).astype(int),
        )

    @classmethod
    def from_mmdetection(cls, mmdet_results) -> Detections:
        """
        Creates a Detections instance from a
        [mmdetection](https://github.com/open-mmlab/mmdetection) and
        [mmyolo](https://github.com/open-mmlab/mmyolo) inference result.

        Args:
            mmdet_results (mmdet.structures.DetDataSample):
                The output Results instance from MMDetection.

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import cv2
            import supervision as sv
            from mmdet.apis import init_detector, inference_detector

            image = cv2.imread(<SOURCE_IMAGE_PATH>)
            model = init_detector(<CONFIG_PATH>, <WEIGHTS_PATH>, device=<DEVICE>)

            result = inference_detector(model, image)
            detections = sv.Detections.from_mmdetection(result)
            ```
        """

        return cls(
            xyxy=mmdet_results.pred_instances.bboxes.cpu().numpy(),
            confidence=mmdet_results.pred_instances.scores.cpu().numpy(),
            class_id=mmdet_results.pred_instances.labels.cpu().numpy().astype(int),
            mask=mmdet_results.pred_instances.masks.cpu().numpy()
            if "masks" in mmdet_results.pred_instances
            else None,
        )

    @classmethod
    def from_transformers(
        cls, transformers_results: dict, id2label: Optional[Dict[int, str]] = None
    ) -> Detections:
        """
        Creates a Detections instance from object detection or panoptic, semantic
        and instance segmentation
        [Transformer](https://github.com/huggingface/transformers) inference result.

        Args:
            transformers_results (Union[dict, torch.Tensor]):  Inference results from
                your Transformers model. This can be either a dictionary containing
                valuable outputs like `scores`, `labels`, `boxes`, `masks`,
                `segments_info`, and `segmentation`, or a `torch.Tensor` holding a
                segmentation map where values represent class IDs.
            id2label (Optional[Dict[int, str]]): A dictionary mapping class IDs to
                labels, typically part of the `transformers` model configuration. If
                provided, the resulting dictionary will include class names.

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import torch
            import supervision as sv
            from PIL import Image
            from transformers import DetrImageProcessor, DetrForObjectDetection

            processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
            model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")

            image = Image.open(<SOURCE_IMAGE_PATH>)
            inputs = processor(images=image, return_tensors="pt")

            with torch.no_grad():
                outputs = model(**inputs)

            width, height = image.size
            target_size = torch.tensor([[height, width]])
            results = processor.post_process_object_detection(
                outputs=outputs, target_sizes=target_size)[0]

            detections = sv.Detections.from_transformers(
                transformers_results=results,
                id2label=model.config.id2label
            )
            ```
        """

        if (
            transformers_results.__class__.__name__ == "Tensor"
            or "segmentation" in transformers_results
        ):
            return cls(
                **process_transformers_v5_segmentation_result(
                    transformers_results, id2label
                )
            )

        if "masks" in transformers_results or "png_string" in transformers_results:
            return cls(
                **process_transformers_v4_segmentation_result(
                    transformers_results, id2label
                )
            )

        if "boxes" in transformers_results:
            return cls(
                **process_transformers_detection_result(transformers_results, id2label)
            )

        else:
            raise ValueError(
                "The provided Transformers results do not contain any valid fields."
                " Expected fields are 'boxes', 'masks', 'segments_info' or"
                " 'segmentation'."
            )

    @classmethod
    def from_detectron2(cls, detectron2_results: Any) -> Detections:
        """
        Create a Detections object from the
        [Detectron2](https://github.com/facebookresearch/detectron2) inference result.

        Args:
            detectron2_results (Any): The output of a
                Detectron2 model containing instances with prediction data.

        Returns:
            (Detections): A Detections object containing the bounding boxes,
                class IDs, and confidences of the predictions.

        Example:
            ```python
            import cv2
            import supervision as sv
            from detectron2.engine import DefaultPredictor
            from detectron2.config import get_cfg


            image = cv2.imread(<SOURCE_IMAGE_PATH>)
            cfg = get_cfg()
            cfg.merge_from_file(<CONFIG_PATH>)
            cfg.MODEL.WEIGHTS = <WEIGHTS_PATH>
            predictor = DefaultPredictor(cfg)

            result = predictor(image)
            detections = sv.Detections.from_detectron2(result)
            ```
        """

        return cls(
            xyxy=detectron2_results["instances"].pred_boxes.tensor.cpu().numpy(),
            confidence=detectron2_results["instances"].scores.cpu().numpy(),
            mask=detectron2_results["instances"].pred_masks.cpu().numpy()
            if hasattr(detectron2_results["instances"], "pred_masks")
            else None,
            class_id=detectron2_results["instances"]
            .pred_classes.cpu()
            .numpy()
            .astype(int),
        )

    @classmethod
    def from_inference(cls, roboflow_result: Union[dict, Any]) -> Detections:
        """
        Create a `sv.Detections` object from the [Roboflow](https://roboflow.com/)
        API inference result or the [Inference](https://inference.roboflow.com/)
        package results. This method extracts bounding boxes, class IDs,
        confidences, and class names from the Roboflow API result and encapsulates
        them into a Detections object.

        Args:
            roboflow_result (dict, any): The result from the
                Roboflow API or Inference package containing predictions.

        Returns:
            (Detections): A Detections object containing the bounding boxes, class IDs,
                and confidences of the predictions.

        Example:
            ```python
            import cv2
            import supervision as sv
            from inference import get_model

            image = cv2.imread(<SOURCE_IMAGE_PATH>)
            model = get_model(model_id="yolov8s-640")

            result = model.infer(image)[0]
            detections = sv.Detections.from_inference(result)
            ```
        """
        if hasattr(roboflow_result, "dict"):
            roboflow_result = roboflow_result.dict(exclude_none=True, by_alias=True)
        elif hasattr(roboflow_result, "json"):
            roboflow_result = roboflow_result.json()
        xyxy, confidence, class_id, masks, trackers, data = process_roboflow_result(
            roboflow_result=roboflow_result
        )

        if np.asarray(xyxy).shape[0] == 0:
            empty_detection = cls.empty()
            empty_detection.data = {CLASS_NAME_DATA_FIELD: np.empty(0)}
            return empty_detection

        return cls(
            xyxy=xyxy,
            confidence=confidence,
            class_id=class_id,
            mask=masks,
            tracker_id=trackers,
            data=data,
        )

    @classmethod
    def from_sam(cls, sam_result: List[dict]) -> Detections:
        """
        Creates a Detections instance from
        [Segment Anything Model](https://github.com/facebookresearch/segment-anything)
        inference result.

        Args:
            sam_result (List[dict]): The output Results instance from SAM

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import supervision as sv
            from segment_anything import (
                sam_model_registry,
                SamAutomaticMaskGenerator
             )

            sam_model_reg = sam_model_registry[MODEL_TYPE]
            sam = sam_model_reg(checkpoint=CHECKPOINT_PATH).to(device=DEVICE)
            mask_generator = SamAutomaticMaskGenerator(sam)
            sam_result = mask_generator.generate(IMAGE)
            detections = sv.Detections.from_sam(sam_result=sam_result)
            ```
        """

        sorted_generated_masks = sorted(
            sam_result, key=lambda x: x["area"], reverse=True
        )

        xywh = np.array([mask["bbox"] for mask in sorted_generated_masks])
        mask = np.array([mask["segmentation"] for mask in sorted_generated_masks])

        if np.asarray(xywh).shape[0] == 0:
            return cls.empty()

        xyxy = xywh_to_xyxy(xywh=xywh)
        return cls(xyxy=xyxy, mask=mask)

    @classmethod
    def from_azure_analyze_image(
        cls, azure_result: dict, class_map: Optional[Dict[int, str]] = None
    ) -> Detections:
        """
        Creates a Detections instance from [Azure Image Analysis 4.0](
        https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/
        concept-object-detection-40).

        Args:
            azure_result (dict): The result from Azure Image Analysis. It should
                contain detected objects and their bounding box coordinates.
            class_map (Optional[Dict[int, str]]): A mapping ofclass IDs (int) to class
                names (str). If None, a new mapping is created dynamically.

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import requests
            import supervision as sv

            image = open(input, "rb").read()

            endpoint = "https://.cognitiveservices.azure.com/"
            subscription_key = ""

            headers = {
                "Content-Type": "application/octet-stream",
                "Ocp-Apim-Subscription-Key": subscription_key
             }

            response = requests.post(endpoint,
                headers=self.headers,
                data=image
             ).json()

            detections = sv.Detections.from_azure_analyze_image(response)
            ```
        """
        if "error" in azure_result:
            raise ValueError(
                f'Azure API returned an error {azure_result["error"]["message"]}'
            )

        xyxy, confidences, class_ids = [], [], []

        is_dynamic_mapping = class_map is None
        if is_dynamic_mapping:
            class_map = {}

        class_map = {value: key for key, value in class_map.items()}

        for detection in azure_result["objectsResult"]["values"]:
            bbox = detection["boundingBox"]

            tags = detection["tags"]

            x0 = bbox["x"]
            y0 = bbox["y"]
            x1 = x0 + bbox["w"]
            y1 = y0 + bbox["h"]

            for tag in tags:
                confidence = tag["confidence"]
                class_name = tag["name"]
                class_id = class_map.get(class_name, None)

                if is_dynamic_mapping and class_id is None:
                    class_id = len(class_map)
                    class_map[class_name] = class_id

                if class_id is not None:
                    xyxy.append([x0, y0, x1, y1])
                    confidences.append(confidence)
                    class_ids.append(class_id)

        if len(xyxy) == 0:
            return Detections.empty()

        return cls(
            xyxy=np.array(xyxy),
            class_id=np.array(class_ids),
            confidence=np.array(confidences),
        )

    @classmethod
    def from_paddledet(cls, paddledet_result) -> Detections:
        """
        Creates a Detections instance from
            [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)
            inference result.

        Args:
            paddledet_result (List[dict]): The output Results instance from PaddleDet

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import supervision as sv
            import paddle
            from ppdet.engine import Trainer
            from ppdet.core.workspace import load_config

            weights = ()
            config = ()

            cfg = load_config(config)
            trainer = Trainer(cfg, mode='test')
            trainer.load_weights(weights)

            paddledet_result = trainer.predict([images])[0]

            detections = sv.Detections.from_paddledet(paddledet_result)
            ```
        """

        if np.asarray(paddledet_result["bbox"][:, 2:6]).shape[0] == 0:
            return cls.empty()

        return cls(
            xyxy=paddledet_result["bbox"][:, 2:6],
            confidence=paddledet_result["bbox"][:, 1],
            class_id=paddledet_result["bbox"][:, 0].astype(int),
        )

    @classmethod
    def from_lmm(
        cls, lmm: Union[LMM, str], result: Union[str, dict], **kwargs: Any
    ) -> Detections:
        """
        Creates a Detections object from the given result string based on the specified
        Large Multimodal Model (LMM).

        Args:
            lmm (Union[LMM, str]): The type of LMM (Large Multimodal Model) to use.
            result (str): The result string containing the detection data.
            **kwargs (Any): Additional keyword arguments required by the specified LMM.

        Returns:
            Detections: A new Detections object.

        Raises:
            ValueError: If the LMM is invalid, required arguments are missing, or
                disallowed arguments are provided.
            ValueError: If the specified LMM is not supported.

        Examples:
            ```python
            import supervision as sv

            paligemma_result = "<loc0256><loc0256><loc0768><loc0768> cat"
            detections = sv.Detections.from_lmm(
                sv.LMM.PALIGEMMA,
                paligemma_result,
                resolution_wh=(1000, 1000),
                classes=['cat', 'dog']
            )
            detections.xyxy
            # array([[250., 250., 750., 750.]])

            detections.class_id
            # array([0])
            ```
        """
        lmm = validate_lmm_parameters(lmm, result, kwargs)

        if lmm == LMM.PALIGEMMA:
            assert isinstance(result, str)
            xyxy, class_id, class_name = from_paligemma(result, **kwargs)
            data = {CLASS_NAME_DATA_FIELD: class_name}
            return cls(xyxy=xyxy, class_id=class_id, data=data)

        if lmm == LMM.FLORENCE_2:
            assert isinstance(result, dict)
            xyxy, labels, mask, xyxyxyxy = from_florence_2(result, **kwargs)
            if len(xyxy) == 0:
                return cls.empty()

            data = {}
            if labels is not None:
                data[CLASS_NAME_DATA_FIELD] = labels
            if xyxyxyxy is not None:
                data[ORIENTED_BOX_COORDINATES] = xyxyxyxy

            return cls(xyxy=xyxy, mask=mask, data=data)

        raise ValueError(f"Unsupported LMM: {lmm}")

    @classmethod
    def from_easyocr(cls, easyocr_results: list) -> Detections:
        """
        Create a Detections object from the
        [EasyOCR](https://github.com/JaidedAI/EasyOCR) result.

        Results are placed in the `data` field with the key `"class_name"`.

        Args:
            easyocr_results (List): The output Results instance from EasyOCR

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import supervision as sv
            import easyocr

            reader = easyocr.Reader(['en'])
            results = reader.readtext(<SOURCE_IMAGE_PATH>)
            detections = sv.Detections.from_easyocr(results)
            detected_text = detections["class_name"]
            ```
        """
        if len(easyocr_results) == 0:
            return cls.empty()

        bbox = np.array([result[0] for result in easyocr_results])
        xyxy = np.hstack((np.min(bbox, axis=1), np.max(bbox, axis=1)))
        confidence = np.array(
            [
                result[2] if len(result) > 2 and result[2] else 0
                for result in easyocr_results
            ]
        )
        ocr_text = np.array([result[1] for result in easyocr_results])

        return cls(
            xyxy=xyxy.astype(np.float32),
            confidence=confidence.astype(np.float32),
            data={
                CLASS_NAME_DATA_FIELD: ocr_text,
            },
        )

    @classmethod
    def from_ncnn(cls, ncnn_results) -> Detections:
        """
        Creates a Detections instance from the
        [ncnn](https://github.com/Tencent/ncnn) inference result.
        Supports object detection models.

        Arguments:
            ncnn_results (dict): The output Results instance from ncnn.

        Returns:
            Detections: A new Detections object.

        Example:
            ```python
            import cv2
            from ncnn.model_zoo import get_model
            import supervision as sv

            image = cv2.imread(<SOURCE_IMAGE_PATH>)
            model = get_model(
                "yolov8s",
                target_size=640
                prob_threshold=0.5,
                nms_threshold=0.45,
                num_threads=4,
                use_gpu=True,
                )
            result = model(image)
            detections = sv.Detections.from_ncnn(result)
            ```
        """

        xywh, confidences, class_ids = [], [], []

        if len(ncnn_results) == 0:
            return cls.empty()

        for ncnn_result in ncnn_results:
            rect = ncnn_result.rect
            xywh.append(
                [
                    rect.x.astype(np.float32),
                    rect.y.astype(np.float32),
                    rect.w.astype(np.float32),
                    rect.h.astype(np.float32),
                ]
            )

            confidences.append(ncnn_result.prob)
            class_ids.append(ncnn_result.label)

        return cls(
            xyxy=xywh_to_xyxy(np.array(xywh, dtype=np.float32)),
            confidence=np.array(confidences, dtype=np.float32),
            class_id=np.array(class_ids, dtype=int),
        )

    @classmethod
    def empty(cls) -> Detections:
        """
        Create an empty Detections object with no bounding boxes,
            confidences, or class IDs.

        Returns:
            (Detections): An empty Detections object.

        Example:
            ```python
            from supervision import Detections

            empty_detections = Detections.empty()
            ```
        """
        return cls(
            xyxy=np.empty((0, 4), dtype=np.float32),
            confidence=np.array([], dtype=np.float32),
            class_id=np.array([], dtype=int),
        )

    def is_empty(self) -> bool:
        """
        Returns `True` if the `Detections` object is considered empty.
        """
        empty_detections = Detections.empty()
        empty_detections.data = self.data
        empty_detections.metadata = self.metadata
        return self == empty_detections

    @classmethod
    def merge(cls, detections_list: List[Detections]) -> Detections:
        """
        Merge a list of Detections objects into a single Detections object.

        This method takes a list of Detections objects and combines their
        respective fields (`xyxy`, `mask`, `confidence`, `class_id`, and `tracker_id`)
        into a single Detections object.

        For example, if merging Detections with 3 and 4 detected objects, this method
        will return a Detections with 7 objects (7 entries in `xyxy`, `mask`, etc).

        !!! Note

            When merging, empty `Detections` objects are ignored.

        Args:
            detections_list (List[Detections]): A list of Detections objects to merge.

        Returns:
            (Detections): A single Detections object containing
                the merged data from the input list.

        Example:
            ```python
            import numpy as np
            import supervision as sv

            detections_1 = sv.Detections(
                xyxy=np.array([[15, 15, 100, 100], [200, 200, 300, 300]]),
                class_id=np.array([1, 2]),
                data={'feature_vector': np.array([0.1, 0.2)])}
             )

            detections_2 = sv.Detections(
                xyxy=np.array([[30, 30, 120, 120]]),
                class_id=np.array([1]),
                data={'feature_vector': [np.array([0.3])]}
             )

            merged_detections = Detections.merge([detections_1, detections_2])

            merged_detections.xyxy
            array([[ 15,  15, 100, 100],
                   [200, 200, 300, 300],
                   [ 30,  30, 120, 120]])

            merged_detections.class_id
            array([1, 2, 1])

            merged_detections.data['feature_vector']
            array([0.1, 0.2, 0.3])
            ```
        """
        detections_list = [
            detections for detections in detections_list if not detections.is_empty()
        ]

        if len(detections_list) == 0:
            return Detections.empty()

        for detections in detections_list:
            validate_detections_fields(
                xyxy=detections.xyxy,
                mask=detections.mask,
                confidence=detections.confidence,
                class_id=detections.class_id,
                tracker_id=detections.tracker_id,
                data=detections.data,
            )

        xyxy = np.vstack([d.xyxy for d in detections_list])

        def stack_or_none(name: str):
            if all(d.__getattribute__(name) is None for d in detections_list):
                return None
            if any(d.__getattribute__(name) is None for d in detections_list):
                raise ValueError(f"All or none of the '{name}' fields must be None")
            return (
                np.vstack([d.__getattribute__(name) for d in detections_list])
                if name == "mask"
                else np.hstack([d.__getattribute__(name) for d in detections_list])
            )

        mask = stack_or_none("mask")
        confidence = stack_or_none("confidence")
        class_id = stack_or_none("class_id")
        tracker_id = stack_or_none("tracker_id")

        data = merge_data([d.data for d in detections_list])

        metadata_list = [detections.metadata for detections in detections_list]
        metadata = merge_metadata(metadata_list)

        return cls(
            xyxy=xyxy,
            mask=mask,
            confidence=confidence,
            class_id=class_id,
            tracker_id=tracker_id,
            data=data,
            metadata=metadata,
        )

    def get_anchors_coordinates(self, anchor: Position) -> np.ndarray:
        """
        Calculates and returns the coordinates of a specific anchor point
        within the bounding boxes defined by the `xyxy` attribute. The anchor
        point can be any of the predefined positions in the `Position` enum,
        such as `CENTER`, `CENTER_LEFT`, `BOTTOM_RIGHT`, etc.

        Args:
            anchor (Position): An enum specifying the position of the anchor point
                within the bounding box. Supported positions are defined in the
                `Position` enum.

        Returns:
            np.ndarray: An array of shape `(n, 2)`, where `n` is the number of bounding
                boxes. Each row contains the `[x, y]` coordinates of the specified
                anchor point for the corresponding bounding box.

        Raises:
            ValueError: If the provided `anchor` is not supported.
        """
        if anchor == Position.CENTER:
            return np.array(
                [
                    (self.xyxy[:, 0] + self.xyxy[:, 2]) / 2,
                    (self.xyxy[:, 1] + self.xyxy[:, 3]) / 2,
                ]
            ).transpose()
        elif anchor == Position.CENTER_OF_MASS:
            if self.mask is None:
                raise ValueError(
                    "Cannot use `Position.CENTER_OF_MASS` without a detection mask."
                )
            return calculate_masks_centroids(masks=self.mask)
        elif anchor == Position.CENTER_LEFT:
            return np.array(
                [
                    self.xyxy[:, 0],
                    (self.xyxy[:, 1] + self.xyxy[:, 3]) / 2,
                ]
            ).transpose()
        elif anchor == Position.CENTER_RIGHT:
            return np.array(
                [
                    self.xyxy[:, 2],
                    (self.xyxy[:, 1] + self.xyxy[:, 3]) / 2,
                ]
            ).transpose()
        elif anchor == Position.BOTTOM_CENTER:
            return np.array(
                [(self.xyxy[:, 0] + self.xyxy[:, 2]) / 2, self.xyxy[:, 3]]
            ).transpose()
        elif anchor == Position.BOTTOM_LEFT:
            return np.array([self.xyxy[:, 0], self.xyxy[:, 3]]).transpose()
        elif anchor == Position.BOTTOM_RIGHT:
            return np.array([self.xyxy[:, 2], self.xyxy[:, 3]]).transpose()
        elif anchor == Position.TOP_CENTER:
            return np.array(
                [(self.xyxy[:, 0] + self.xyxy[:, 2]) / 2, self.xyxy[:, 1]]
            ).transpose()
        elif anchor == Position.TOP_LEFT:
            return np.array([self.xyxy[:, 0], self.xyxy[:, 1]]).transpose()
        elif anchor == Position.TOP_RIGHT:
            return np.array([self.xyxy[:, 2], self.xyxy[:, 1]]).transpose()

        raise ValueError(f"{anchor} is not supported.")

    def __getitem__(
        self, index: Union[int, slice, List[int], np.ndarray, str]
    ) -> Union[Detections, List, np.ndarray, None]:
        """
        Get a subset of the Detections object or access an item from its data field.

        When provided with an integer, slice, list of integers, or a numpy array, this
        method returns a new Detections object that represents a subset of the original
        detections. When provided with a string, it accesses the corresponding item in
        the data dictionary.

        Args:
            index (Union[int, slice, List[int], np.ndarray, str]): The index, indices,
                or key to access a subset of the Detections or an item from the data.

        Returns:
            Union[Detections, Any]: A subset of the Detections object or an item from
                the data field.

        Example:
            ```python
            import supervision as sv

            detections = sv.Detections()

            first_detection = detections[0]
            first_10_detections = detections[0:10]
            some_detections = detections[[0, 2, 4]]
            class_0_detections = detections[detections.class_id == 0]
            high_confidence_detections = detections[detections.confidence > 0.5]

            feature_vector = detections['feature_vector']
            ```
        """
        if isinstance(index, str):
            return self.data.get(index)
        if self.is_empty():
            return Detections.empty()
        if isinstance(index, int):
            index = [index]
        return Detections(
            xyxy=self.xyxy[index],
            mask=self.mask[index] if self.mask is not None else None,
            confidence=self.confidence[index] if self.confidence is not None else None,
            class_id=self.class_id[index] if self.class_id is not None else None,
            tracker_id=self.tracker_id[index] if self.tracker_id is not None else None,
            data=get_data_item(self.data, index),
            metadata=self.metadata,
        )

    def __setitem__(self, key: str, value: Union[np.ndarray, List]):
        """
        Set a value in the data dictionary of the Detections object.

        Args:
            key (str): The key in the data dictionary to set.
            value (Union[np.ndarray, List]): The value to set for the key.

        Example:
            ```python
            import cv2
            import supervision as sv
            from ultralytics import YOLO

            image = cv2.imread(<SOURCE_IMAGE_PATH>)
            model = YOLO('yolov8s.pt')

            result = model(image)[0]
            detections = sv.Detections.from_ultralytics(result)

            detections['names'] = [
                 model.model.names[class_id]
                 for class_id
                 in detections.class_id
             ]
            ```
        """
        if not isinstance(value, (np.ndarray, list)):
            raise TypeError("Value must be a np.ndarray or a list")

        if isinstance(value, list):
            value = np.array(value)

        self.data[key] = value

    @property
    def area(self) -> np.ndarray:
        """
        Calculate the area of each detection in the set of object detections.
        If masks field is defined property returns are of each mask.
        If only box is given property return area of each box.

        Returns:
          np.ndarray: An array of floats containing the area of each detection
            in the format of `(area_1, area_2, , area_n)`,
            where n is the number of detections.
        """
        if self.mask is not None:
            return np.array([np.sum(mask) for mask in self.mask])
        else:
            return self.box_area

    @property
    def box_area(self) -> np.ndarray:
        """
        Calculate the area of each bounding box in the set of object detections.

        Returns:
            np.ndarray: An array of floats containing the area of each bounding
                box in the format of `(area_1, area_2, , area_n)`,
                where n is the number of detections.
        """
        return (self.xyxy[:, 3] - self.xyxy[:, 1]) * (self.xyxy[:, 2] - self.xyxy[:, 0])

    def with_nms(
        self, threshold: float = 0.5, class_agnostic: bool = False
    ) -> Detections:
        """
        Performs non-max suppression on detection set. If the detections result
        from a segmentation model, the IoU mask is applied. Otherwise, box IoU is used.

        Args:
            threshold (float): The intersection-over-union threshold
                to use for non-maximum suppression. I'm the lower the value the more
                restrictive the NMS becomes. Defaults to 0.5.
            class_agnostic (bool): Whether to perform class-agnostic
                non-maximum suppression. If True, the class_id of each detection
                will be ignored. Defaults to False.

        Returns:
            Detections: A new Detections object containing the subset of detections
                after non-maximum suppression.

        Raises:
            AssertionError: If `confidence` is None and class_agnostic is False.
                If `class_id` is None and class_agnostic is False.
        """
        if len(self) == 0:
            return self

        assert (
            self.confidence is not None
        ), "Detections confidence must be given for NMS to be executed."

        if class_agnostic:
            predictions = np.hstack((self.xyxy, self.confidence.reshape(-1, 1)))
        else:
            assert self.class_id is not None, (
                "Detections class_id must be given for NMS to be executed. If you"
                " intended to perform class agnostic NMS set class_agnostic=True."
            )
            predictions = np.hstack(
                (
                    self.xyxy,
                    self.confidence.reshape(-1, 1),
                    self.class_id.reshape(-1, 1),
                )
            )

        if self.mask is not None:
            indices = mask_non_max_suppression(
                predictions=predictions, masks=self.mask, iou_threshold=threshold
            )
        else:
            indices = box_non_max_suppression(
                predictions=predictions, iou_threshold=threshold
            )

        return self[indices]

    def with_nmm(
        self, threshold: float = 0.5, class_agnostic: bool = False
    ) -> Detections:
        """
        Perform non-maximum merging on the current set of object detections.

        Args:
            threshold (float): The intersection-over-union threshold
                to use for non-maximum merging. Defaults to 0.5.
            class_agnostic (bool): Whether to perform class-agnostic
                non-maximum merging. If True, the class_id of each detection
                will be ignored. Defaults to False.

        Returns:
            Detections: A new Detections object containing the subset of detections
                after non-maximum merging.

        Raises:
            AssertionError: If `confidence` is None or `class_id` is None and
                class_agnostic is False.

        ![non-max-merging](https://media.roboflow.com/supervision-docs/non-max-merging.png){ align=center width="800" }
        """  # noqa: E501 // docs
        if len(self) == 0:
            return self

        assert (
            self.confidence is not None
        ), "Detections confidence must be given for NMM to be executed."

        if class_agnostic:
            predictions = np.hstack((self.xyxy, self.confidence.reshape(-1, 1)))
        else:
            assert self.class_id is not None, (
                "Detections class_id must be given for NMM to be executed. If you"
                " intended to perform class agnostic NMM set class_agnostic=True."
            )
            predictions = np.hstack(
                (
                    self.xyxy,
                    self.confidence.reshape(-1, 1),
                    self.class_id.reshape(-1, 1),
                )
            )

        merge_groups = box_non_max_merge(
            predictions=predictions, iou_threshold=threshold
        )

        result = []
        for merge_group in merge_groups:
            unmerged_detections = [self[i] for i in merge_group]
            merged_detections = merge_inner_detections_objects(
                unmerged_detections, threshold
            )
            result.append(merged_detections)

        return Detections.merge(result)

Attributes

area: np.ndarray property

Calculate the area of each detection in the set of object detections. If masks field is defined property returns are of each mask. If only box is given property return area of each box.

Returns:

Type Description
ndarray

np.ndarray: An array of floats containing the area of each detection in the format of (area_1, area_2, , area_n), where n is the number of detections.

box_area: np.ndarray property

Calculate the area of each bounding box in the set of object detections.

Returns:

Type Description
ndarray

np.ndarray: An array of floats containing the area of each bounding box in the format of (area_1, area_2, , area_n), where n is the number of detections.

Functions

__getitem__(index)

Get a subset of the Detections object or access an item from its data field.

When provided with an integer, slice, list of integers, or a numpy array, this method returns a new Detections object that represents a subset of the original detections. When provided with a string, it accesses the corresponding item in the data dictionary.

Parameters:

Name Type Description Default

index

Union[int, slice, List[int], ndarray, str]

The index, indices, or key to access a subset of the Detections or an item from the data.

required

Returns:

Type Description
Union[Detections, List, ndarray, None]

Union[Detections, Any]: A subset of the Detections object or an item from the data field.

Example
import supervision as sv

detections = sv.Detections()

first_detection = detections[0]
first_10_detections = detections[0:10]
some_detections = detections[[0, 2, 4]]
class_0_detections = detections[detections.class_id == 0]
high_confidence_detections = detections[detections.confidence > 0.5]

feature_vector = detections['feature_vector']
Source code in supervision/detection/core.py
def __getitem__(
    self, index: Union[int, slice, List[int], np.ndarray, str]
) -> Union[Detections, List, np.ndarray, None]:
    """
    Get a subset of the Detections object or access an item from its data field.

    When provided with an integer, slice, list of integers, or a numpy array, this
    method returns a new Detections object that represents a subset of the original
    detections. When provided with a string, it accesses the corresponding item in
    the data dictionary.

    Args:
        index (Union[int, slice, List[int], np.ndarray, str]): The index, indices,
            or key to access a subset of the Detections or an item from the data.

    Returns:
        Union[Detections, Any]: A subset of the Detections object or an item from
            the data field.

    Example:
        ```python
        import supervision as sv

        detections = sv.Detections()

        first_detection = detections[0]
        first_10_detections = detections[0:10]
        some_detections = detections[[0, 2, 4]]
        class_0_detections = detections[detections.class_id == 0]
        high_confidence_detections = detections[detections.confidence > 0.5]

        feature_vector = detections['feature_vector']
        ```
    """
    if isinstance(index, str):
        return self.data.get(index)
    if self.is_empty():
        return Detections.empty()
    if isinstance(index, int):
        index = [index]
    return Detections(
        xyxy=self.xyxy[index],
        mask=self.mask[index] if self.mask is not None else None,
        confidence=self.confidence[index] if self.confidence is not None else None,
        class_id=self.class_id[index] if self.class_id is not None else None,
        tracker_id=self.tracker_id[index] if self.tracker_id is not None else None,
        data=get_data_item(self.data, index),
        metadata=self.metadata,
    )

__iter__()

Iterates over the Detections object and yield a tuple of (xyxy, mask, confidence, class_id, tracker_id, data) for each detection.

Source code in supervision/detection/core.py
def __iter__(
    self,
) -> Iterator[
    Tuple[
        np.ndarray,
        Optional[np.ndarray],
        Optional[float],
        Optional[int],
        Optional[int],
        Dict[str, Union[np.ndarray, List]],
    ]
]:
    """
    Iterates over the Detections object and yield a tuple of
    `(xyxy, mask, confidence, class_id, tracker_id, data)` for each detection.
    """
    for i in range(len(self.xyxy)):
        yield (
            self.xyxy[i],
            self.mask[i] if self.mask is not None else None,
            self.confidence[i] if self.confidence is not None else None,
            self.class_id[i] if self.class_id is not None else None,
            self.tracker_id[i] if self.tracker_id is not None else None,
            get_data_item(self.data, i),
        )

__len__()

Returns the number of detections in the Detections object.

Source code in supervision/detection/core.py
def __len__(self):
    """
    Returns the number of detections in the Detections object.
    """
    return len(self.xyxy)

__setitem__(key, value)

Set a value in the data dictionary of the Detections object.

Parameters:

Name Type Description Default

key

str

The key in the data dictionary to set.

required

value

Union[ndarray, List]

The value to set for the key.

required
Example
import cv2
import supervision as sv
from ultralytics import YOLO

image = cv2.imread(<SOURCE_IMAGE_PATH>)
model = YOLO('yolov8s.pt')

result = model(image)[0]
detections = sv.Detections.from_ultralytics(result)

detections['names'] = [
     model.model.names[class_id]
     for class_id
     in detections.class_id
 ]
Source code in supervision/detection/core.py
def __setitem__(self, key: str, value: Union[np.ndarray, List]):
    """
    Set a value in the data dictionary of the Detections object.

    Args:
        key (str): The key in the data dictionary to set.
        value (Union[np.ndarray, List]): The value to set for the key.

    Example:
        ```python
        import cv2
        import supervision as sv
        from ultralytics import YOLO

        image = cv2.imread(<SOURCE_IMAGE_PATH>)
        model = YOLO('yolov8s.pt')

        result = model(image)[0]
        detections = sv.Detections.from_ultralytics(result)

        detections['names'] = [
             model.model.names[class_id]
             for class_id
             in detections.class_id
         ]
        ```
    """
    if not isinstance(value, (np.ndarray, list)):
        raise TypeError("Value must be a np.ndarray or a list")

    if isinstance(value, list):
        value = np.array(value)

    self.data[key] = value

empty() classmethod

Create an empty Detections object with no bounding boxes, confidences, or class IDs.

Returns:

Type Description
Detections

An empty Detections object.

Example
from supervision import Detections

empty_detections = Detections.empty()
Source code in supervision/detection/core.py
@classmethod
def empty(cls) -> Detections:
    """
    Create an empty Detections object with no bounding boxes,
        confidences, or class IDs.

    Returns:
        (Detections): An empty Detections object.

    Example:
        ```python
        from supervision import Detections

        empty_detections = Detections.empty()
        ```
    """
    return cls(
        xyxy=np.empty((0, 4), dtype=np.float32),
        confidence=np.array([], dtype=np.float32),
        class_id=np.array([], dtype=int),
    )

from_azure_analyze_image(azure_result, class_map=None) classmethod

Creates a Detections instance from Azure Image Analysis 4.0.

Parameters:

Name Type Description Default

azure_result

dict

The result from Azure Image Analysis. It should contain detected objects and their bounding box coordinates.

required

class_map

Optional[Dict[int, str]]

A mapping ofclass IDs (int) to class names (str). If None, a new mapping is created dynamically.

None

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import requests
import supervision as sv

image = open(input, "rb").read()

endpoint = "https://.cognitiveservices.azure.com/"
subscription_key = ""

headers = {
    "Content-Type": "application/octet-stream",
    "Ocp-Apim-Subscription-Key": subscription_key
 }

response = requests.post(endpoint,
    headers=self.headers,
    data=image
 ).json()

detections = sv.Detections.from_azure_analyze_image(response)
Source code in supervision/detection/core.py
@classmethod
def from_azure_analyze_image(
    cls, azure_result: dict, class_map: Optional[Dict[int, str]] = None
) -> Detections:
    """
    Creates a Detections instance from [Azure Image Analysis 4.0](
    https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/
    concept-object-detection-40).

    Args:
        azure_result (dict): The result from Azure Image Analysis. It should
            contain detected objects and their bounding box coordinates.
        class_map (Optional[Dict[int, str]]): A mapping ofclass IDs (int) to class
            names (str). If None, a new mapping is created dynamically.

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import requests
        import supervision as sv

        image = open(input, "rb").read()

        endpoint = "https://.cognitiveservices.azure.com/"
        subscription_key = ""

        headers = {
            "Content-Type": "application/octet-stream",
            "Ocp-Apim-Subscription-Key": subscription_key
         }

        response = requests.post(endpoint,
            headers=self.headers,
            data=image
         ).json()

        detections = sv.Detections.from_azure_analyze_image(response)
        ```
    """
    if "error" in azure_result:
        raise ValueError(
            f'Azure API returned an error {azure_result["error"]["message"]}'
        )

    xyxy, confidences, class_ids = [], [], []

    is_dynamic_mapping = class_map is None
    if is_dynamic_mapping:
        class_map = {}

    class_map = {value: key for key, value in class_map.items()}

    for detection in azure_result["objectsResult"]["values"]:
        bbox = detection["boundingBox"]

        tags = detection["tags"]

        x0 = bbox["x"]
        y0 = bbox["y"]
        x1 = x0 + bbox["w"]
        y1 = y0 + bbox["h"]

        for tag in tags:
            confidence = tag["confidence"]
            class_name = tag["name"]
            class_id = class_map.get(class_name, None)

            if is_dynamic_mapping and class_id is None:
                class_id = len(class_map)
                class_map[class_name] = class_id

            if class_id is not None:
                xyxy.append([x0, y0, x1, y1])
                confidences.append(confidence)
                class_ids.append(class_id)

    if len(xyxy) == 0:
        return Detections.empty()

    return cls(
        xyxy=np.array(xyxy),
        class_id=np.array(class_ids),
        confidence=np.array(confidences),
    )

from_deepsparse(deepsparse_results) classmethod

Creates a Detections instance from a DeepSparse inference result.

Parameters:

Name Type Description Default

deepsparse_results

YOLOOutput

The output Results instance from DeepSparse.

required

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import supervision as sv
from deepsparse import Pipeline

yolo_pipeline = Pipeline.create(
    task="yolo",
    model_path = "zoo:cv/detection/yolov5-l/pytorch/ultralytics/coco/pruned80_quant-none"
 )
result = yolo_pipeline(<SOURCE IMAGE PATH>)
detections = sv.Detections.from_deepsparse(result)
Source code in supervision/detection/core.py
@classmethod
def from_deepsparse(cls, deepsparse_results) -> Detections:
    """
    Creates a Detections instance from a
    [DeepSparse](https://github.com/neuralmagic/deepsparse)
    inference result.

    Args:
        deepsparse_results (deepsparse.yolo.schemas.YOLOOutput):
            The output Results instance from DeepSparse.

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import supervision as sv
        from deepsparse import Pipeline

        yolo_pipeline = Pipeline.create(
            task="yolo",
            model_path = "zoo:cv/detection/yolov5-l/pytorch/ultralytics/coco/pruned80_quant-none"
         )
        result = yolo_pipeline(<SOURCE IMAGE PATH>)
        detections = sv.Detections.from_deepsparse(result)
        ```
    """  # noqa: E501 // docs

    if np.asarray(deepsparse_results.boxes[0]).shape[0] == 0:
        return cls.empty()

    return cls(
        xyxy=np.array(deepsparse_results.boxes[0]),
        confidence=np.array(deepsparse_results.scores[0]),
        class_id=np.array(deepsparse_results.labels[0]).astype(float).astype(int),
    )

from_detectron2(detectron2_results) classmethod

Create a Detections object from the Detectron2 inference result.

Parameters:

Name Type Description Default

detectron2_results

Any

The output of a Detectron2 model containing instances with prediction data.

required

Returns:

Type Description
Detections

A Detections object containing the bounding boxes, class IDs, and confidences of the predictions.

Example
import cv2
import supervision as sv
from detectron2.engine import DefaultPredictor
from detectron2.config import get_cfg


image = cv2.imread(<SOURCE_IMAGE_PATH>)
cfg = get_cfg()
cfg.merge_from_file(<CONFIG_PATH>)
cfg.MODEL.WEIGHTS = <WEIGHTS_PATH>
predictor = DefaultPredictor(cfg)

result = predictor(image)
detections = sv.Detections.from_detectron2(result)
Source code in supervision/detection/core.py
@classmethod
def from_detectron2(cls, detectron2_results: Any) -> Detections:
    """
    Create a Detections object from the
    [Detectron2](https://github.com/facebookresearch/detectron2) inference result.

    Args:
        detectron2_results (Any): The output of a
            Detectron2 model containing instances with prediction data.

    Returns:
        (Detections): A Detections object containing the bounding boxes,
            class IDs, and confidences of the predictions.

    Example:
        ```python
        import cv2
        import supervision as sv
        from detectron2.engine import DefaultPredictor
        from detectron2.config import get_cfg


        image = cv2.imread(<SOURCE_IMAGE_PATH>)
        cfg = get_cfg()
        cfg.merge_from_file(<CONFIG_PATH>)
        cfg.MODEL.WEIGHTS = <WEIGHTS_PATH>
        predictor = DefaultPredictor(cfg)

        result = predictor(image)
        detections = sv.Detections.from_detectron2(result)
        ```
    """

    return cls(
        xyxy=detectron2_results["instances"].pred_boxes.tensor.cpu().numpy(),
        confidence=detectron2_results["instances"].scores.cpu().numpy(),
        mask=detectron2_results["instances"].pred_masks.cpu().numpy()
        if hasattr(detectron2_results["instances"], "pred_masks")
        else None,
        class_id=detectron2_results["instances"]
        .pred_classes.cpu()
        .numpy()
        .astype(int),
    )

from_easyocr(easyocr_results) classmethod

Create a Detections object from the EasyOCR result.

Results are placed in the data field with the key "class_name".

Parameters:

Name Type Description Default

easyocr_results

List

The output Results instance from EasyOCR

required

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import supervision as sv
import easyocr

reader = easyocr.Reader(['en'])
results = reader.readtext(<SOURCE_IMAGE_PATH>)
detections = sv.Detections.from_easyocr(results)
detected_text = detections["class_name"]
Source code in supervision/detection/core.py
@classmethod
def from_easyocr(cls, easyocr_results: list) -> Detections:
    """
    Create a Detections object from the
    [EasyOCR](https://github.com/JaidedAI/EasyOCR) result.

    Results are placed in the `data` field with the key `"class_name"`.

    Args:
        easyocr_results (List): The output Results instance from EasyOCR

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import supervision as sv
        import easyocr

        reader = easyocr.Reader(['en'])
        results = reader.readtext(<SOURCE_IMAGE_PATH>)
        detections = sv.Detections.from_easyocr(results)
        detected_text = detections["class_name"]
        ```
    """
    if len(easyocr_results) == 0:
        return cls.empty()

    bbox = np.array([result[0] for result in easyocr_results])
    xyxy = np.hstack((np.min(bbox, axis=1), np.max(bbox, axis=1)))
    confidence = np.array(
        [
            result[2] if len(result) > 2 and result[2] else 0
            for result in easyocr_results
        ]
    )
    ocr_text = np.array([result[1] for result in easyocr_results])

    return cls(
        xyxy=xyxy.astype(np.float32),
        confidence=confidence.astype(np.float32),
        data={
            CLASS_NAME_DATA_FIELD: ocr_text,
        },
    )

from_inference(roboflow_result) classmethod

Create a sv.Detections object from the Roboflow API inference result or the Inference package results. This method extracts bounding boxes, class IDs, confidences, and class names from the Roboflow API result and encapsulates them into a Detections object.

Parameters:

Name Type Description Default

roboflow_result

(dict, any)

The result from the Roboflow API or Inference package containing predictions.

required

Returns:

Type Description
Detections

A Detections object containing the bounding boxes, class IDs, and confidences of the predictions.

Example
import cv2
import supervision as sv
from inference import get_model

image = cv2.imread(<SOURCE_IMAGE_PATH>)
model = get_model(model_id="yolov8s-640")

result = model.infer(image)[0]
detections = sv.Detections.from_inference(result)
Source code in supervision/detection/core.py
@classmethod
def from_inference(cls, roboflow_result: Union[dict, Any]) -> Detections:
    """
    Create a `sv.Detections` object from the [Roboflow](https://roboflow.com/)
    API inference result or the [Inference](https://inference.roboflow.com/)
    package results. This method extracts bounding boxes, class IDs,
    confidences, and class names from the Roboflow API result and encapsulates
    them into a Detections object.

    Args:
        roboflow_result (dict, any): The result from the
            Roboflow API or Inference package containing predictions.

    Returns:
        (Detections): A Detections object containing the bounding boxes, class IDs,
            and confidences of the predictions.

    Example:
        ```python
        import cv2
        import supervision as sv
        from inference import get_model

        image = cv2.imread(<SOURCE_IMAGE_PATH>)
        model = get_model(model_id="yolov8s-640")

        result = model.infer(image)[0]
        detections = sv.Detections.from_inference(result)
        ```
    """
    if hasattr(roboflow_result, "dict"):
        roboflow_result = roboflow_result.dict(exclude_none=True, by_alias=True)
    elif hasattr(roboflow_result, "json"):
        roboflow_result = roboflow_result.json()
    xyxy, confidence, class_id, masks, trackers, data = process_roboflow_result(
        roboflow_result=roboflow_result
    )

    if np.asarray(xyxy).shape[0] == 0:
        empty_detection = cls.empty()
        empty_detection.data = {CLASS_NAME_DATA_FIELD: np.empty(0)}
        return empty_detection

    return cls(
        xyxy=xyxy,
        confidence=confidence,
        class_id=class_id,
        mask=masks,
        tracker_id=trackers,
        data=data,
    )

from_lmm(lmm, result, **kwargs) classmethod

Creates a Detections object from the given result string based on the specified Large Multimodal Model (LMM).

Parameters:

Name Type Description Default

lmm

Union[LMM, str]

The type of LMM (Large Multimodal Model) to use.

required

result

str

The result string containing the detection data.

required

**kwargs

Any

Additional keyword arguments required by the specified LMM.

{}

Returns:

Name Type Description
Detections Detections

A new Detections object.

Raises:

Type Description
ValueError

If the LMM is invalid, required arguments are missing, or disallowed arguments are provided.

ValueError

If the specified LMM is not supported.

Examples:

import supervision as sv

paligemma_result = "<loc0256><loc0256><loc0768><loc0768> cat"
detections = sv.Detections.from_lmm(
    sv.LMM.PALIGEMMA,
    paligemma_result,
    resolution_wh=(1000, 1000),
    classes=['cat', 'dog']
)
detections.xyxy
# array([[250., 250., 750., 750.]])

detections.class_id
# array([0])
Source code in supervision/detection/core.py
@classmethod
def from_lmm(
    cls, lmm: Union[LMM, str], result: Union[str, dict], **kwargs: Any
) -> Detections:
    """
    Creates a Detections object from the given result string based on the specified
    Large Multimodal Model (LMM).

    Args:
        lmm (Union[LMM, str]): The type of LMM (Large Multimodal Model) to use.
        result (str): The result string containing the detection data.
        **kwargs (Any): Additional keyword arguments required by the specified LMM.

    Returns:
        Detections: A new Detections object.

    Raises:
        ValueError: If the LMM is invalid, required arguments are missing, or
            disallowed arguments are provided.
        ValueError: If the specified LMM is not supported.

    Examples:
        ```python
        import supervision as sv

        paligemma_result = "<loc0256><loc0256><loc0768><loc0768> cat"
        detections = sv.Detections.from_lmm(
            sv.LMM.PALIGEMMA,
            paligemma_result,
            resolution_wh=(1000, 1000),
            classes=['cat', 'dog']
        )
        detections.xyxy
        # array([[250., 250., 750., 750.]])

        detections.class_id
        # array([0])
        ```
    """
    lmm = validate_lmm_parameters(lmm, result, kwargs)

    if lmm == LMM.PALIGEMMA:
        assert isinstance(result, str)
        xyxy, class_id, class_name = from_paligemma(result, **kwargs)
        data = {CLASS_NAME_DATA_FIELD: class_name}
        return cls(xyxy=xyxy, class_id=class_id, data=data)

    if lmm == LMM.FLORENCE_2:
        assert isinstance(result, dict)
        xyxy, labels, mask, xyxyxyxy = from_florence_2(result, **kwargs)
        if len(xyxy) == 0:
            return cls.empty()

        data = {}
        if labels is not None:
            data[CLASS_NAME_DATA_FIELD] = labels
        if xyxyxyxy is not None:
            data[ORIENTED_BOX_COORDINATES] = xyxyxyxy

        return cls(xyxy=xyxy, mask=mask, data=data)

    raise ValueError(f"Unsupported LMM: {lmm}")

from_mmdetection(mmdet_results) classmethod

Creates a Detections instance from a mmdetection and mmyolo inference result.

Parameters:

Name Type Description Default

mmdet_results

DetDataSample

The output Results instance from MMDetection.

required

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import cv2
import supervision as sv
from mmdet.apis import init_detector, inference_detector

image = cv2.imread(<SOURCE_IMAGE_PATH>)
model = init_detector(<CONFIG_PATH>, <WEIGHTS_PATH>, device=<DEVICE>)

result = inference_detector(model, image)
detections = sv.Detections.from_mmdetection(result)
Source code in supervision/detection/core.py
@classmethod
def from_mmdetection(cls, mmdet_results) -> Detections:
    """
    Creates a Detections instance from a
    [mmdetection](https://github.com/open-mmlab/mmdetection) and
    [mmyolo](https://github.com/open-mmlab/mmyolo) inference result.

    Args:
        mmdet_results (mmdet.structures.DetDataSample):
            The output Results instance from MMDetection.

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import cv2
        import supervision as sv
        from mmdet.apis import init_detector, inference_detector

        image = cv2.imread(<SOURCE_IMAGE_PATH>)
        model = init_detector(<CONFIG_PATH>, <WEIGHTS_PATH>, device=<DEVICE>)

        result = inference_detector(model, image)
        detections = sv.Detections.from_mmdetection(result)
        ```
    """

    return cls(
        xyxy=mmdet_results.pred_instances.bboxes.cpu().numpy(),
        confidence=mmdet_results.pred_instances.scores.cpu().numpy(),
        class_id=mmdet_results.pred_instances.labels.cpu().numpy().astype(int),
        mask=mmdet_results.pred_instances.masks.cpu().numpy()
        if "masks" in mmdet_results.pred_instances
        else None,
    )

from_ncnn(ncnn_results) classmethod

Creates a Detections instance from the ncnn inference result. Supports object detection models.

Parameters:

Name Type Description Default

ncnn_results

dict

The output Results instance from ncnn.

required

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import cv2
from ncnn.model_zoo import get_model
import supervision as sv

image = cv2.imread(<SOURCE_IMAGE_PATH>)
model = get_model(
    "yolov8s",
    target_size=640
    prob_threshold=0.5,
    nms_threshold=0.45,
    num_threads=4,
    use_gpu=True,
    )
result = model(image)
detections = sv.Detections.from_ncnn(result)
Source code in supervision/detection/core.py
@classmethod
def from_ncnn(cls, ncnn_results) -> Detections:
    """
    Creates a Detections instance from the
    [ncnn](https://github.com/Tencent/ncnn) inference result.
    Supports object detection models.

    Arguments:
        ncnn_results (dict): The output Results instance from ncnn.

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import cv2
        from ncnn.model_zoo import get_model
        import supervision as sv

        image = cv2.imread(<SOURCE_IMAGE_PATH>)
        model = get_model(
            "yolov8s",
            target_size=640
            prob_threshold=0.5,
            nms_threshold=0.45,
            num_threads=4,
            use_gpu=True,
            )
        result = model(image)
        detections = sv.Detections.from_ncnn(result)
        ```
    """

    xywh, confidences, class_ids = [], [], []

    if len(ncnn_results) == 0:
        return cls.empty()

    for ncnn_result in ncnn_results:
        rect = ncnn_result.rect
        xywh.append(
            [
                rect.x.astype(np.float32),
                rect.y.astype(np.float32),
                rect.w.astype(np.float32),
                rect.h.astype(np.float32),
            ]
        )

        confidences.append(ncnn_result.prob)
        class_ids.append(ncnn_result.label)

    return cls(
        xyxy=xywh_to_xyxy(np.array(xywh, dtype=np.float32)),
        confidence=np.array(confidences, dtype=np.float32),
        class_id=np.array(class_ids, dtype=int),
    )

from_paddledet(paddledet_result) classmethod

Creates a Detections instance from PaddleDetection inference result.

Parameters:

Name Type Description Default

paddledet_result

List[dict]

The output Results instance from PaddleDet

required

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import supervision as sv
import paddle
from ppdet.engine import Trainer
from ppdet.core.workspace import load_config

weights = ()
config = ()

cfg = load_config(config)
trainer = Trainer(cfg, mode='test')
trainer.load_weights(weights)

paddledet_result = trainer.predict([images])[0]

detections = sv.Detections.from_paddledet(paddledet_result)
Source code in supervision/detection/core.py
@classmethod
def from_paddledet(cls, paddledet_result) -> Detections:
    """
    Creates a Detections instance from
        [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)
        inference result.

    Args:
        paddledet_result (List[dict]): The output Results instance from PaddleDet

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import supervision as sv
        import paddle
        from ppdet.engine import Trainer
        from ppdet.core.workspace import load_config

        weights = ()
        config = ()

        cfg = load_config(config)
        trainer = Trainer(cfg, mode='test')
        trainer.load_weights(weights)

        paddledet_result = trainer.predict([images])[0]

        detections = sv.Detections.from_paddledet(paddledet_result)
        ```
    """

    if np.asarray(paddledet_result["bbox"][:, 2:6]).shape[0] == 0:
        return cls.empty()

    return cls(
        xyxy=paddledet_result["bbox"][:, 2:6],
        confidence=paddledet_result["bbox"][:, 1],
        class_id=paddledet_result["bbox"][:, 0].astype(int),
    )

from_sam(sam_result) classmethod

Creates a Detections instance from Segment Anything Model inference result.

Parameters:

Name Type Description Default

sam_result

List[dict]

The output Results instance from SAM

required

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import supervision as sv
from segment_anything import (
    sam_model_registry,
    SamAutomaticMaskGenerator
 )

sam_model_reg = sam_model_registry[MODEL_TYPE]
sam = sam_model_reg(checkpoint=CHECKPOINT_PATH).to(device=DEVICE)
mask_generator = SamAutomaticMaskGenerator(sam)
sam_result = mask_generator.generate(IMAGE)
detections = sv.Detections.from_sam(sam_result=sam_result)
Source code in supervision/detection/core.py
@classmethod
def from_sam(cls, sam_result: List[dict]) -> Detections:
    """
    Creates a Detections instance from
    [Segment Anything Model](https://github.com/facebookresearch/segment-anything)
    inference result.

    Args:
        sam_result (List[dict]): The output Results instance from SAM

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import supervision as sv
        from segment_anything import (
            sam_model_registry,
            SamAutomaticMaskGenerator
         )

        sam_model_reg = sam_model_registry[MODEL_TYPE]
        sam = sam_model_reg(checkpoint=CHECKPOINT_PATH).to(device=DEVICE)
        mask_generator = SamAutomaticMaskGenerator(sam)
        sam_result = mask_generator.generate(IMAGE)
        detections = sv.Detections.from_sam(sam_result=sam_result)
        ```
    """

    sorted_generated_masks = sorted(
        sam_result, key=lambda x: x["area"], reverse=True
    )

    xywh = np.array([mask["bbox"] for mask in sorted_generated_masks])
    mask = np.array([mask["segmentation"] for mask in sorted_generated_masks])

    if np.asarray(xywh).shape[0] == 0:
        return cls.empty()

    xyxy = xywh_to_xyxy(xywh=xywh)
    return cls(xyxy=xyxy, mask=mask)

from_tensorflow(tensorflow_results, resolution_wh) classmethod

Creates a Detections instance from a Tensorflow Hub inference result.

Parameters:

Name Type Description Default

tensorflow_results

dict

The output results from Tensorflow Hub.

required

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import tensorflow as tf
import tensorflow_hub as hub
import numpy as np
import cv2

module_handle = "https://tfhub.dev/tensorflow/centernet/hourglass_512x512_kpts/1"
model = hub.load(module_handle)
img = np.array(cv2.imread(SOURCE_IMAGE_PATH))
result = model(img)
detections = sv.Detections.from_tensorflow(result)
Source code in supervision/detection/core.py
@classmethod
def from_tensorflow(
    cls, tensorflow_results: dict, resolution_wh: tuple
) -> Detections:
    """
    Creates a Detections instance from a
    [Tensorflow Hub](https://www.tensorflow.org/hub/tutorials/tf2_object_detection)
    inference result.

    Args:
        tensorflow_results (dict):
            The output results from Tensorflow Hub.

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import tensorflow as tf
        import tensorflow_hub as hub
        import numpy as np
        import cv2

        module_handle = "https://tfhub.dev/tensorflow/centernet/hourglass_512x512_kpts/1"
        model = hub.load(module_handle)
        img = np.array(cv2.imread(SOURCE_IMAGE_PATH))
        result = model(img)
        detections = sv.Detections.from_tensorflow(result)
        ```
    """

    boxes = tensorflow_results["detection_boxes"][0].numpy()
    boxes[:, [0, 2]] *= resolution_wh[0]
    boxes[:, [1, 3]] *= resolution_wh[1]
    boxes = boxes[:, [1, 0, 3, 2]]
    return cls(
        xyxy=boxes,
        confidence=tensorflow_results["detection_scores"][0].numpy(),
        class_id=tensorflow_results["detection_classes"][0].numpy().astype(int),
    )

from_transformers(transformers_results, id2label=None) classmethod

Creates a Detections instance from object detection or panoptic, semantic and instance segmentation Transformer inference result.

Parameters:

Name Type Description Default

transformers_results

Union[dict, Tensor]

Inference results from your Transformers model. This can be either a dictionary containing valuable outputs like scores, labels, boxes, masks, segments_info, and segmentation, or a torch.Tensor holding a segmentation map where values represent class IDs.

required

id2label

Optional[Dict[int, str]]

A dictionary mapping class IDs to labels, typically part of the transformers model configuration. If provided, the resulting dictionary will include class names.

None

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import torch
import supervision as sv
from PIL import Image
from transformers import DetrImageProcessor, DetrForObjectDetection

processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")

image = Image.open(<SOURCE_IMAGE_PATH>)
inputs = processor(images=image, return_tensors="pt")

with torch.no_grad():
    outputs = model(**inputs)

width, height = image.size
target_size = torch.tensor([[height, width]])
results = processor.post_process_object_detection(
    outputs=outputs, target_sizes=target_size)[0]

detections = sv.Detections.from_transformers(
    transformers_results=results,
    id2label=model.config.id2label
)
Source code in supervision/detection/core.py
@classmethod
def from_transformers(
    cls, transformers_results: dict, id2label: Optional[Dict[int, str]] = None
) -> Detections:
    """
    Creates a Detections instance from object detection or panoptic, semantic
    and instance segmentation
    [Transformer](https://github.com/huggingface/transformers) inference result.

    Args:
        transformers_results (Union[dict, torch.Tensor]):  Inference results from
            your Transformers model. This can be either a dictionary containing
            valuable outputs like `scores`, `labels`, `boxes`, `masks`,
            `segments_info`, and `segmentation`, or a `torch.Tensor` holding a
            segmentation map where values represent class IDs.
        id2label (Optional[Dict[int, str]]): A dictionary mapping class IDs to
            labels, typically part of the `transformers` model configuration. If
            provided, the resulting dictionary will include class names.

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import torch
        import supervision as sv
        from PIL import Image
        from transformers import DetrImageProcessor, DetrForObjectDetection

        processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
        model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")

        image = Image.open(<SOURCE_IMAGE_PATH>)
        inputs = processor(images=image, return_tensors="pt")

        with torch.no_grad():
            outputs = model(**inputs)

        width, height = image.size
        target_size = torch.tensor([[height, width]])
        results = processor.post_process_object_detection(
            outputs=outputs, target_sizes=target_size)[0]

        detections = sv.Detections.from_transformers(
            transformers_results=results,
            id2label=model.config.id2label
        )
        ```
    """

    if (
        transformers_results.__class__.__name__ == "Tensor"
        or "segmentation" in transformers_results
    ):
        return cls(
            **process_transformers_v5_segmentation_result(
                transformers_results, id2label
            )
        )

    if "masks" in transformers_results or "png_string" in transformers_results:
        return cls(
            **process_transformers_v4_segmentation_result(
                transformers_results, id2label
            )
        )

    if "boxes" in transformers_results:
        return cls(
            **process_transformers_detection_result(transformers_results, id2label)
        )

    else:
        raise ValueError(
            "The provided Transformers results do not contain any valid fields."
            " Expected fields are 'boxes', 'masks', 'segments_info' or"
            " 'segmentation'."
        )

from_ultralytics(ultralytics_results) classmethod

Creates a sv.Detections instance from a YOLOv8 inference result.

Note

from_ultralytics is compatible with detection, segmentation, and OBB models.

Parameters:

Name Type Description Default

ultralytics_results

Results

The output Results instance from Ultralytics

required

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import cv2
import supervision as sv
from ultralytics import YOLO

image = cv2.imread(<SOURCE_IMAGE_PATH>)
model = YOLO('yolov8s.pt')
results = model(image)[0]
detections = sv.Detections.from_ultralytics(results)
Source code in supervision/detection/core.py
@classmethod
def from_ultralytics(cls, ultralytics_results) -> Detections:
    """
    Creates a `sv.Detections` instance from a
    [YOLOv8](https://github.com/ultralytics/ultralytics) inference result.

    !!! Note

        `from_ultralytics` is compatible with
        [detection](https://docs.ultralytics.com/tasks/detect/),
        [segmentation](https://docs.ultralytics.com/tasks/segment/), and
        [OBB](https://docs.ultralytics.com/tasks/obb/) models.

    Args:
        ultralytics_results (ultralytics.yolo.engine.results.Results):
            The output Results instance from Ultralytics

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import cv2
        import supervision as sv
        from ultralytics import YOLO

        image = cv2.imread(<SOURCE_IMAGE_PATH>)
        model = YOLO('yolov8s.pt')
        results = model(image)[0]
        detections = sv.Detections.from_ultralytics(results)
        ```
    """

    if hasattr(ultralytics_results, "obb") and ultralytics_results.obb is not None:
        class_id = ultralytics_results.obb.cls.cpu().numpy().astype(int)
        class_names = np.array([ultralytics_results.names[i] for i in class_id])
        oriented_box_coordinates = ultralytics_results.obb.xyxyxyxy.cpu().numpy()
        return cls(
            xyxy=ultralytics_results.obb.xyxy.cpu().numpy(),
            confidence=ultralytics_results.obb.conf.cpu().numpy(),
            class_id=class_id,
            tracker_id=ultralytics_results.obb.id.int().cpu().numpy()
            if ultralytics_results.obb.id is not None
            else None,
            data={
                ORIENTED_BOX_COORDINATES: oriented_box_coordinates,
                CLASS_NAME_DATA_FIELD: class_names,
            },
        )

    if hasattr(ultralytics_results, "boxes") and ultralytics_results.boxes is None:
        masks = extract_ultralytics_masks(ultralytics_results)
        return cls(
            xyxy=mask_to_xyxy(masks),
            mask=masks,
            class_id=np.arange(len(ultralytics_results)),
        )

    class_id = ultralytics_results.boxes.cls.cpu().numpy().astype(int)
    class_names = np.array([ultralytics_results.names[i] for i in class_id])
    return cls(
        xyxy=ultralytics_results.boxes.xyxy.cpu().numpy(),
        confidence=ultralytics_results.boxes.conf.cpu().numpy(),
        class_id=class_id,
        mask=extract_ultralytics_masks(ultralytics_results),
        tracker_id=ultralytics_results.boxes.id.int().cpu().numpy()
        if ultralytics_results.boxes.id is not None
        else None,
        data={CLASS_NAME_DATA_FIELD: class_names},
    )

from_yolo_nas(yolo_nas_results) classmethod

Creates a Detections instance from a YOLO-NAS inference result.

Parameters:

Name Type Description Default

yolo_nas_results

ImageDetectionPrediction

The output Results instance from YOLO-NAS ImageDetectionPrediction is coming from 'super_gradients.training.models.prediction_results'

required

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import cv2
from super_gradients.training import models
import supervision as sv

image = cv2.imread(<SOURCE_IMAGE_PATH>)
model = models.get('yolo_nas_l', pretrained_weights="coco")

result = list(model.predict(image, conf=0.35))[0]
detections = sv.Detections.from_yolo_nas(result)
Source code in supervision/detection/core.py
@classmethod
def from_yolo_nas(cls, yolo_nas_results) -> Detections:
    """
    Creates a Detections instance from a
    [YOLO-NAS](https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS.md)
    inference result.

    Args:
        yolo_nas_results (ImageDetectionPrediction):
            The output Results instance from YOLO-NAS
            ImageDetectionPrediction is coming from
            'super_gradients.training.models.prediction_results'

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import cv2
        from super_gradients.training import models
        import supervision as sv

        image = cv2.imread(<SOURCE_IMAGE_PATH>)
        model = models.get('yolo_nas_l', pretrained_weights="coco")

        result = list(model.predict(image, conf=0.35))[0]
        detections = sv.Detections.from_yolo_nas(result)
        ```
    """
    if np.asarray(yolo_nas_results.prediction.bboxes_xyxy).shape[0] == 0:
        return cls.empty()

    return cls(
        xyxy=yolo_nas_results.prediction.bboxes_xyxy,
        confidence=yolo_nas_results.prediction.confidence,
        class_id=yolo_nas_results.prediction.labels.astype(int),
    )

from_yolov5(yolov5_results) classmethod

Creates a Detections instance from a YOLOv5 inference result.

Parameters:

Name Type Description Default

yolov5_results

Detections

The output Detections instance from YOLOv5

required

Returns:

Name Type Description
Detections Detections

A new Detections object.

Example
import cv2
import torch
import supervision as sv

image = cv2.imread(<SOURCE_IMAGE_PATH>)
model = torch.hub.load('ultralytics/yolov5', 'yolov5s')
result = model(image)
detections = sv.Detections.from_yolov5(result)
Source code in supervision/detection/core.py
@classmethod
def from_yolov5(cls, yolov5_results) -> Detections:
    """
    Creates a Detections instance from a
    [YOLOv5](https://github.com/ultralytics/yolov5) inference result.

    Args:
        yolov5_results (yolov5.models.common.Detections):
            The output Detections instance from YOLOv5

    Returns:
        Detections: A new Detections object.

    Example:
        ```python
        import cv2
        import torch
        import supervision as sv

        image = cv2.imread(<SOURCE_IMAGE_PATH>)
        model = torch.hub.load('ultralytics/yolov5', 'yolov5s')
        result = model(image)
        detections = sv.Detections.from_yolov5(result)
        ```
    """
    yolov5_detections_predictions = yolov5_results.pred[0].cpu().cpu().numpy()

    return cls(
        xyxy=yolov5_detections_predictions[:, :4],
        confidence=yolov5_detections_predictions[:, 4],
        class_id=yolov5_detections_predictions[:, 5].astype(int),
    )

get_anchors_coordinates(anchor)

Calculates and returns the coordinates of a specific anchor point within the bounding boxes defined by the xyxy attribute. The anchor point can be any of the predefined positions in the Position enum, such as CENTER, CENTER_LEFT, BOTTOM_RIGHT, etc.

Parameters:

Name Type Description Default

anchor

Position

An enum specifying the position of the anchor point within the bounding box. Supported positions are defined in the Position enum.

required

Returns:

Type Description
ndarray

np.ndarray: An array of shape (n, 2), where n is the number of bounding boxes. Each row contains the [x, y] coordinates of the specified anchor point for the corresponding bounding box.

Raises:

Type Description
ValueError

If the provided anchor is not supported.

Source code in supervision/detection/core.py
def get_anchors_coordinates(self, anchor: Position) -> np.ndarray:
    """
    Calculates and returns the coordinates of a specific anchor point
    within the bounding boxes defined by the `xyxy` attribute. The anchor
    point can be any of the predefined positions in the `Position` enum,
    such as `CENTER`, `CENTER_LEFT`, `BOTTOM_RIGHT`, etc.

    Args:
        anchor (Position): An enum specifying the position of the anchor point
            within the bounding box. Supported positions are defined in the
            `Position` enum.

    Returns:
        np.ndarray: An array of shape `(n, 2)`, where `n` is the number of bounding
            boxes. Each row contains the `[x, y]` coordinates of the specified
            anchor point for the corresponding bounding box.

    Raises:
        ValueError: If the provided `anchor` is not supported.
    """
    if anchor == Position.CENTER:
        return np.array(
            [
                (self.xyxy[:, 0] + self.xyxy[:, 2]) / 2,
                (self.xyxy[:, 1] + self.xyxy[:, 3]) / 2,
            ]
        ).transpose()
    elif anchor == Position.CENTER_OF_MASS:
        if self.mask is None:
            raise ValueError(
                "Cannot use `Position.CENTER_OF_MASS` without a detection mask."
            )
        return calculate_masks_centroids(masks=self.mask)
    elif anchor == Position.CENTER_LEFT:
        return np.array(
            [
                self.xyxy[:, 0],
                (self.xyxy[:, 1] + self.xyxy[:, 3]) / 2,
            ]
        ).transpose()
    elif anchor == Position.CENTER_RIGHT:
        return np.array(
            [
                self.xyxy[:, 2],
                (self.xyxy[:, 1] + self.xyxy[:, 3]) / 2,
            ]
        ).transpose()
    elif anchor == Position.BOTTOM_CENTER:
        return np.array(
            [(self.xyxy[:, 0] + self.xyxy[:, 2]) / 2, self.xyxy[:, 3]]
        ).transpose()
    elif anchor == Position.BOTTOM_LEFT:
        return np.array([self.xyxy[:, 0], self.xyxy[:, 3]]).transpose()
    elif anchor == Position.BOTTOM_RIGHT:
        return np.array([self.xyxy[:, 2], self.xyxy[:, 3]]).transpose()
    elif anchor == Position.TOP_CENTER:
        return np.array(
            [(self.xyxy[:, 0] + self.xyxy[:, 2]) / 2, self.xyxy[:, 1]]
        ).transpose()
    elif anchor == Position.TOP_LEFT:
        return np.array([self.xyxy[:, 0], self.xyxy[:, 1]]).transpose()
    elif anchor == Position.TOP_RIGHT:
        return np.array([self.xyxy[:, 2], self.xyxy[:, 1]]).transpose()

    raise ValueError(f"{anchor} is not supported.")

is_empty()

Returns True if the Detections object is considered empty.

Source code in supervision/detection/core.py
def is_empty(self) -> bool:
    """
    Returns `True` if the `Detections` object is considered empty.
    """
    empty_detections = Detections.empty()
    empty_detections.data = self.data
    empty_detections.metadata = self.metadata
    return self == empty_detections

merge(detections_list) classmethod

Merge a list of Detections objects into a single Detections object.

This method takes a list of Detections objects and combines their respective fields (xyxy, mask, confidence, class_id, and tracker_id) into a single Detections object.

For example, if merging Detections with 3 and 4 detected objects, this method will return a Detections with 7 objects (7 entries in xyxy, mask, etc).

Note

When merging, empty Detections objects are ignored.

Parameters:

Name Type Description Default

detections_list

List[Detections]

A list of Detections objects to merge.

required

Returns:

Type Description
Detections

A single Detections object containing the merged data from the input list.

Example
import numpy as np
import supervision as sv

detections_1 = sv.Detections(
    xyxy=np.array([[15, 15, 100, 100], [200, 200, 300, 300]]),
    class_id=np.array([1, 2]),
    data={'feature_vector': np.array([0.1, 0.2)])}
 )

detections_2 = sv.Detections(
    xyxy=np.array([[30, 30, 120, 120]]),
    class_id=np.array([1]),
    data={'feature_vector': [np.array([0.3])]}
 )

merged_detections = Detections.merge([detections_1, detections_2])

merged_detections.xyxy
array([[ 15,  15, 100, 100],
       [200, 200, 300, 300],
       [ 30,  30, 120, 120]])

merged_detections.class_id
array([1, 2, 1])

merged_detections.data['feature_vector']
array([0.1, 0.2, 0.3])
Source code in supervision/detection/core.py
@classmethod
def merge(cls, detections_list: List[Detections]) -> Detections:
    """
    Merge a list of Detections objects into a single Detections object.

    This method takes a list of Detections objects and combines their
    respective fields (`xyxy`, `mask`, `confidence`, `class_id`, and `tracker_id`)
    into a single Detections object.

    For example, if merging Detections with 3 and 4 detected objects, this method
    will return a Detections with 7 objects (7 entries in `xyxy`, `mask`, etc).

    !!! Note

        When merging, empty `Detections` objects are ignored.

    Args:
        detections_list (List[Detections]): A list of Detections objects to merge.

    Returns:
        (Detections): A single Detections object containing
            the merged data from the input list.

    Example:
        ```python
        import numpy as np
        import supervision as sv

        detections_1 = sv.Detections(
            xyxy=np.array([[15, 15, 100, 100], [200, 200, 300, 300]]),
            class_id=np.array([1, 2]),
            data={'feature_vector': np.array([0.1, 0.2)])}
         )

        detections_2 = sv.Detections(
            xyxy=np.array([[30, 30, 120, 120]]),
            class_id=np.array([1]),
            data={'feature_vector': [np.array([0.3])]}
         )

        merged_detections = Detections.merge([detections_1, detections_2])

        merged_detections.xyxy
        array([[ 15,  15, 100, 100],
               [200, 200, 300, 300],
               [ 30,  30, 120, 120]])

        merged_detections.class_id
        array([1, 2, 1])

        merged_detections.data['feature_vector']
        array([0.1, 0.2, 0.3])
        ```
    """
    detections_list = [
        detections for detections in detections_list if not detections.is_empty()
    ]

    if len(detections_list) == 0:
        return Detections.empty()

    for detections in detections_list:
        validate_detections_fields(
            xyxy=detections.xyxy,
            mask=detections.mask,
            confidence=detections.confidence,
            class_id=detections.class_id,
            tracker_id=detections.tracker_id,
            data=detections.data,
        )

    xyxy = np.vstack([d.xyxy for d in detections_list])

    def stack_or_none(name: str):
        if all(d.__getattribute__(name) is None for d in detections_list):
            return None
        if any(d.__getattribute__(name) is None for d in detections_list):
            raise ValueError(f"All or none of the '{name}' fields must be None")
        return (
            np.vstack([d.__getattribute__(name) for d in detections_list])
            if name == "mask"
            else np.hstack([d.__getattribute__(name) for d in detections_list])
        )

    mask = stack_or_none("mask")
    confidence = stack_or_none("confidence")
    class_id = stack_or_none("class_id")
    tracker_id = stack_or_none("tracker_id")

    data = merge_data([d.data for d in detections_list])

    metadata_list = [detections.metadata for detections in detections_list]
    metadata = merge_metadata(metadata_list)

    return cls(
        xyxy=xyxy,
        mask=mask,
        confidence=confidence,
        class_id=class_id,
        tracker_id=tracker_id,
        data=data,
        metadata=metadata,
    )

with_nmm(threshold=0.5, class_agnostic=False)

Perform non-maximum merging on the current set of object detections.

Parameters:

Name Type Description Default

threshold

float

The intersection-over-union threshold to use for non-maximum merging. Defaults to 0.5.

0.5

class_agnostic

bool

Whether to perform class-agnostic non-maximum merging. If True, the class_id of each detection will be ignored. Defaults to False.

False

Returns:

Name Type Description
Detections Detections

A new Detections object containing the subset of detections after non-maximum merging.

Raises:

Type Description
AssertionError

If confidence is None or class_id is None and class_agnostic is False.

non-max-merging

Source code in supervision/detection/core.py
def with_nmm(
    self, threshold: float = 0.5, class_agnostic: bool = False
) -> Detections:
    """
    Perform non-maximum merging on the current set of object detections.

    Args:
        threshold (float): The intersection-over-union threshold
            to use for non-maximum merging. Defaults to 0.5.
        class_agnostic (bool): Whether to perform class-agnostic
            non-maximum merging. If True, the class_id of each detection
            will be ignored. Defaults to False.

    Returns:
        Detections: A new Detections object containing the subset of detections
            after non-maximum merging.

    Raises:
        AssertionError: If `confidence` is None or `class_id` is None and
            class_agnostic is False.

    ![non-max-merging](https://media.roboflow.com/supervision-docs/non-max-merging.png){ align=center width="800" }
    """  # noqa: E501 // docs
    if len(self) == 0:
        return self

    assert (
        self.confidence is not None
    ), "Detections confidence must be given for NMM to be executed."

    if class_agnostic:
        predictions = np.hstack((self.xyxy, self.confidence.reshape(-1, 1)))
    else:
        assert self.class_id is not None, (
            "Detections class_id must be given for NMM to be executed. If you"
            " intended to perform class agnostic NMM set class_agnostic=True."
        )
        predictions = np.hstack(
            (
                self.xyxy,
                self.confidence.reshape(-1, 1),
                self.class_id.reshape(-1, 1),
            )
        )

    merge_groups = box_non_max_merge(
        predictions=predictions, iou_threshold=threshold
    )

    result = []
    for merge_group in merge_groups:
        unmerged_detections = [self[i] for i in merge_group]
        merged_detections = merge_inner_detections_objects(
            unmerged_detections, threshold
        )
        result.append(merged_detections)

    return Detections.merge(result)

with_nms(threshold=0.5, class_agnostic=False)

Performs non-max suppression on detection set. If the detections result from a segmentation model, the IoU mask is applied. Otherwise, box IoU is used.

Parameters:

Name Type Description Default

threshold

float

The intersection-over-union threshold to use for non-maximum suppression. I'm the lower the value the more restrictive the NMS becomes. Defaults to 0.5.

0.5

class_agnostic

bool

Whether to perform class-agnostic non-maximum suppression. If True, the class_id of each detection will be ignored. Defaults to False.

False

Returns:

Name Type Description
Detections Detections

A new Detections object containing the subset of detections after non-maximum suppression.

Raises:

Type Description
AssertionError

If confidence is None and class_agnostic is False. If class_id is None and class_agnostic is False.

Source code in supervision/detection/core.py
def with_nms(
    self, threshold: float = 0.5, class_agnostic: bool = False
) -> Detections:
    """
    Performs non-max suppression on detection set. If the detections result
    from a segmentation model, the IoU mask is applied. Otherwise, box IoU is used.

    Args:
        threshold (float): The intersection-over-union threshold
            to use for non-maximum suppression. I'm the lower the value the more
            restrictive the NMS becomes. Defaults to 0.5.
        class_agnostic (bool): Whether to perform class-agnostic
            non-maximum suppression. If True, the class_id of each detection
            will be ignored. Defaults to False.

    Returns:
        Detections: A new Detections object containing the subset of detections
            after non-maximum suppression.

    Raises:
        AssertionError: If `confidence` is None and class_agnostic is False.
            If `class_id` is None and class_agnostic is False.
    """
    if len(self) == 0:
        return self

    assert (
        self.confidence is not None
    ), "Detections confidence must be given for NMS to be executed."

    if class_agnostic:
        predictions = np.hstack((self.xyxy, self.confidence.reshape(-1, 1)))
    else:
        assert self.class_id is not None, (
            "Detections class_id must be given for NMS to be executed. If you"
            " intended to perform class agnostic NMS set class_agnostic=True."
        )
        predictions = np.hstack(
            (
                self.xyxy,
                self.confidence.reshape(-1, 1),
                self.class_id.reshape(-1, 1),
            )
        )

    if self.mask is not None:
        indices = mask_non_max_suppression(
            predictions=predictions, masks=self.mask, iou_threshold=threshold
        )
    else:
        indices = box_non_max_suppression(
            predictions=predictions, iou_threshold=threshold
        )

    return self[indices]

Comments