Skip to content

Classifications

supervision.classification.core.Classifications dataclass

Source code in src/supervision/classification/core.py 
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 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
@dataclass
class Classifications:
    class_id: npt.NDArray[np.int_]
    confidence: npt.NDArray[np.floating] | None = None

    def __post_init__(self) -> None:
        """
        Validate the classification inputs.
        """
        n = len(self.class_id)

        _validate_class_ids(self.class_id, n)
        _validate_confidence(self.confidence, n)

    def __eq__(self, other: object) -> bool:
        """
        Compare classifications by value across numpy-backed fields.
        """
        if not isinstance(other, Classifications):
            return NotImplemented
        if not np.array_equal(self.class_id, other.class_id):
            return False
        if self.confidence is None or other.confidence is None:
            return self.confidence is other.confidence
        return bool(np.array_equal(self.confidence, other.confidence))

    def __len__(self) -> int:
        """
        Returns the number of classifications.
        """
        return len(self.class_id)

    @classmethod
    def from_clip(cls, clip_results: torch.Tensor) -> Classifications:
        """
        Creates a Classifications instance from a
        [clip](https://github.com/openai/clip) inference result.

        Args:
            clip_results: The inference result from clip model.

        Returns:
            A new Classifications object.

        Example:
            ```python
            from PIL import Image
            import clip
            import supervision as sv

            model, preprocess = clip.load('ViT-B/32')

            image = cv2.imread(SOURCE_IMAGE_PATH)
            image = preprocess(image).unsqueeze(0)

            text = clip.tokenize(["a diagram", "a dog", "a cat"])
            output, _ = model(image, text)
            classifications = sv.Classifications.from_clip(output)
            ```
        """

        confidence = clip_results.softmax(dim=-1).cpu().detach().numpy()[0]

        if len(confidence) == 0:
            return cls(
                class_id=np.array([], dtype=np.int_),
                confidence=np.array([], dtype=np.float32),
            )

        class_ids = np.arange(len(confidence))
        return cls(class_id=class_ids, confidence=confidence)

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

        Args:
            ultralytics_results: The inference result from ultralytics model.

        Returns:
            A new Classifications object.

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

            image = cv2.imread(SOURCE_IMAGE_PATH)
            model = YOLO('yolov8n-cls.pt')

            output = model(image)[0]
            classifications = sv.Classifications.from_ultralytics(output)
            ```
        """
        confidence = ultralytics_results.probs.data.cpu().numpy()
        return cls(class_id=np.arange(confidence.shape[0]), confidence=confidence)

    @classmethod
    def from_timm(cls, timm_results: Any) -> Classifications:
        """
        Creates a Classifications instance from a
        [timm](https://huggingface.co/docs/hub/timm) inference result.

        Note:
            Returned confidences are softmax-normalized probabilities, so
            thresholds calibrated against raw logits may need recalibration.

        Args:
            timm_results: The inference result from timm model.

        Returns:
            A new Classifications object.

        Example:
            ```python
            from PIL import Image
            import timm
            from timm.data import resolve_data_config, create_transform
            import supervision as sv

            model = timm.create_model(
                model_name='hf-hub:nateraw/resnet50-oxford-iiit-pet',
                pretrained=True
            ).eval()

            config = resolve_data_config({}, model=model)
            transform = create_transform(**config)

            image = Image.open(SOURCE_IMAGE_PATH).convert('RGB')
            x = transform(image).unsqueeze(0)

            output = model(x)

            classifications = sv.Classifications.from_timm(output)
            ```
        """
        confidence = timm_results.softmax(dim=-1).cpu().detach().numpy()[0]

        if len(confidence) == 0:
            return cls(
                class_id=np.array([], dtype=np.int_),
                confidence=np.array([], dtype=np.float32),
            )

        class_id = np.arange(len(confidence))
        return cls(class_id=class_id, confidence=confidence)

    def get_top_k(
        self, k: int
    ) -> tuple[npt.NDArray[np.int_], npt.NDArray[np.floating]]:
        """
        Retrieve the top k class IDs and confidences,
            ordered in descending order by confidence.

        Args:
            k: The number of top class IDs and confidences to retrieve.

        Returns:
            A tuple containing the top k class IDs and confidences.

        Example:
            ```pycon
            >>> import numpy as np
            >>> import supervision as sv
            >>> classifications = sv.Classifications(
            ...     class_id=np.array([0, 1, 2]),
            ...     confidence=np.array([0.3, 0.9, 0.5])
            ... )
            >>> classifications.get_top_k(1)
            (array([1]), array([0.9]))

            ```
        """
        if self.confidence is None:
            raise ValueError("top_k could not be calculated, confidence is None")

        order = np.argsort(self.confidence)[::-1]
        top_k_order = order[:k]
        top_k_class_id = self.class_id[top_k_order]
        top_k_confidence = self.confidence[top_k_order]

        return top_k_class_id, top_k_confidence

Methods:

__eq__(other: object) -> bool

Compare classifications by value across numpy-backed fields.

Source code in src/supervision/classification/core.py 
46
47
48
49
50
51
52
53
54
55
56
def __eq__(self, other: object) -> bool:
    """
    Compare classifications by value across numpy-backed fields.
    """
    if not isinstance(other, Classifications):
        return NotImplemented
    if not np.array_equal(self.class_id, other.class_id):
        return False
    if self.confidence is None or other.confidence is None:
        return self.confidence is other.confidence
    return bool(np.array_equal(self.confidence, other.confidence))

__len__() -> int

Returns the number of classifications.

Source code in src/supervision/classification/core.py 
58
59
60
61
62
def __len__(self) -> int:
    """
    Returns the number of classifications.
    """
    return len(self.class_id)

__post_init__() -> None

Validate the classification inputs.

Source code in src/supervision/classification/core.py 
37
38
39
40
41
42
43
44
def __post_init__(self) -> None:
    """
    Validate the classification inputs.
    """
    n = len(self.class_id)

    _validate_class_ids(self.class_id, n)
    _validate_confidence(self.confidence, n)

from_clip(clip_results: torch.Tensor) -> Classifications classmethod

Creates a Classifications instance from a clip inference result.

Parameters:

Name Type Description Default
clip_results
Tensor

The inference result from clip model.

 required

Returns:

Type Description
Classifications

A new Classifications object.
Example 
from PIL import Image
import clip
import supervision as sv

model, preprocess = clip.load('ViT-B/32')

image = cv2.imread(SOURCE_IMAGE_PATH)
image = preprocess(image).unsqueeze(0)

text = clip.tokenize(["a diagram", "a dog", "a cat"])
output, _ = model(image, text)
classifications = sv.Classifications.from_clip(output)
Source code in src/supervision/classification/core.py 
 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
@classmethod
def from_clip(cls, clip_results: torch.Tensor) -> Classifications:
    """
    Creates a Classifications instance from a
    [clip](https://github.com/openai/clip) inference result.

    Args:
        clip_results: The inference result from clip model.

    Returns:
        A new Classifications object.

    Example:
        ```python
        from PIL import Image
        import clip
        import supervision as sv

        model, preprocess = clip.load('ViT-B/32')

        image = cv2.imread(SOURCE_IMAGE_PATH)
        image = preprocess(image).unsqueeze(0)

        text = clip.tokenize(["a diagram", "a dog", "a cat"])
        output, _ = model(image, text)
        classifications = sv.Classifications.from_clip(output)
        ```
    """

    confidence = clip_results.softmax(dim=-1).cpu().detach().numpy()[0]

    if len(confidence) == 0:
        return cls(
            class_id=np.array([], dtype=np.int_),
            confidence=np.array([], dtype=np.float32),
        )

    class_ids = np.arange(len(confidence))
    return cls(class_id=class_ids, confidence=confidence)

from_timm(timm_results: Any) -> Classifications classmethod

Creates a Classifications instance from a timm inference result.

Note

Returned confidences are softmax-normalized probabilities, so thresholds calibrated against raw logits may need recalibration.

Parameters:

Name Type Description Default
timm_results
Any

The inference result from timm model.

 required

Returns:

Type Description
Classifications

A new Classifications object.
Example 
from PIL import Image
import timm
from timm.data import resolve_data_config, create_transform
import supervision as sv

model = timm.create_model(
    model_name='hf-hub:nateraw/resnet50-oxford-iiit-pet',
    pretrained=True
).eval()

config = resolve_data_config({}, model=model)
transform = create_transform(**config)

image = Image.open(SOURCE_IMAGE_PATH).convert('RGB')
x = transform(image).unsqueeze(0)

output = model(x)

classifications = sv.Classifications.from_timm(output)
Source code in src/supervision/classification/core.py 
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
@classmethod
def from_timm(cls, timm_results: Any) -> Classifications:
    """
    Creates a Classifications instance from a
    [timm](https://huggingface.co/docs/hub/timm) inference result.

    Note:
        Returned confidences are softmax-normalized probabilities, so
        thresholds calibrated against raw logits may need recalibration.

    Args:
        timm_results: The inference result from timm model.

    Returns:
        A new Classifications object.

    Example:
        ```python
        from PIL import Image
        import timm
        from timm.data import resolve_data_config, create_transform
        import supervision as sv

        model = timm.create_model(
            model_name='hf-hub:nateraw/resnet50-oxford-iiit-pet',
            pretrained=True
        ).eval()

        config = resolve_data_config({}, model=model)
        transform = create_transform(**config)

        image = Image.open(SOURCE_IMAGE_PATH).convert('RGB')
        x = transform(image).unsqueeze(0)

        output = model(x)

        classifications = sv.Classifications.from_timm(output)
        ```
    """
    confidence = timm_results.softmax(dim=-1).cpu().detach().numpy()[0]

    if len(confidence) == 0:
        return cls(
            class_id=np.array([], dtype=np.int_),
            confidence=np.array([], dtype=np.float32),
        )

    class_id = np.arange(len(confidence))
    return cls(class_id=class_id, confidence=confidence)

from_ultralytics(ultralytics_results: Any) -> Classifications classmethod

Creates a Classifications instance from a ultralytics inference result.

Parameters:

Name Type Description Default
ultralytics_results
Any

The inference result from ultralytics model.

 required

Returns:

Type Description
Classifications

A new Classifications object.
Example 
import cv2
from ultralytics import YOLO
import supervision as sv

image = cv2.imread(SOURCE_IMAGE_PATH)
model = YOLO('yolov8n-cls.pt')

output = model(image)[0]
classifications = sv.Classifications.from_ultralytics(output)
Source code in src/supervision/classification/core.py 
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
@classmethod
def from_ultralytics(cls, ultralytics_results: Any) -> Classifications:
    """
    Creates a Classifications instance from a
    [ultralytics](https://github.com/ultralytics/ultralytics) inference result.

    Args:
        ultralytics_results: The inference result from ultralytics model.

    Returns:
        A new Classifications object.

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

        image = cv2.imread(SOURCE_IMAGE_PATH)
        model = YOLO('yolov8n-cls.pt')

        output = model(image)[0]
        classifications = sv.Classifications.from_ultralytics(output)
        ```
    """
    confidence = ultralytics_results.probs.data.cpu().numpy()
    return cls(class_id=np.arange(confidence.shape[0]), confidence=confidence)

get_top_k(k: int) -> tuple[npt.NDArray[np.int_], npt.NDArray[np.floating]]

Retrieve the top k class IDs and confidences, ordered in descending order by confidence.

Parameters:

Name Type Description Default
k
int

The number of top class IDs and confidences to retrieve.

 required

Returns:

Type Description
tuple[NDArray[int_], NDArray[floating]]

A tuple containing the top k class IDs and confidences.
Example 
>>> import numpy as np
>>> import supervision as sv
>>> classifications = sv.Classifications(
...     class_id=np.array([0, 1, 2]),
...     confidence=np.array([0.3, 0.9, 0.5])
... )
>>> classifications.get_top_k(1)
(array([1]), array([0.9]))
Source code in src/supervision/classification/core.py 
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
def get_top_k(
    self, k: int
) -> tuple[npt.NDArray[np.int_], npt.NDArray[np.floating]]:
    """
    Retrieve the top k class IDs and confidences,
        ordered in descending order by confidence.

    Args:
        k: The number of top class IDs and confidences to retrieve.

    Returns:
        A tuple containing the top k class IDs and confidences.

    Example:
        ```pycon
        >>> import numpy as np
        >>> import supervision as sv
        >>> classifications = sv.Classifications(
        ...     class_id=np.array([0, 1, 2]),
        ...     confidence=np.array([0.3, 0.9, 0.5])
        ... )
        >>> classifications.get_top_k(1)
        (array([1]), array([0.9]))

        ```
    """
    if self.confidence is None:
        raise ValueError("top_k could not be calculated, confidence is None")

    order = np.argsort(self.confidence)[::-1]
    top_k_order = order[:k]
    top_k_class_id = self.class_id[top_k_order]
    top_k_confidence = self.confidence[top_k_order]

    return top_k_class_id, top_k_confidence

Comments