Predictions

Note

The ArrayLike type is expected to be a np.ndarray or a torch.Tensor.

mozuma.predictions.BatchModelPrediction dataclass

Class defining the accepted types for model's predictions

Attributes:

Name	Type	Description
features	ArrayLike | None	Features or embeddings. Dimensions=(dataset_size, feature_dims...)
label_scores	ArrayLike | None	Score for each predicted label. Dimensions=(dataset_size, n_labels)
frames	Sequence[BatchVideoFramesPrediction[ArrayLike]] | None	Prediction for each frame and video. See BatchVideoFramesPrediction. Sequence length=dataset_size
bounding_boxes	Sequence[BatchBoundingBoxesPrediction[ArrayLike]] | None	Prediction for each bounding_box and image. See BatchBoundingBoxesPrediction. Sequence length=dataset_size

mozuma.predictions.BatchBoundingBoxesPrediction dataclass

Results of bounding box extraction on an images

Attributes:

Name	Type	Description
bounding_boxes	ArrayLike	Array of bounding box coordinates. Dimensions=(n_boxes, 4).
scores	ArrayLike | None	Array of scores for each bounding boxes. Dimensions=(n_boxes, 1).
features	ArrayLike | None	Array of features for each bounding boxes. Dimensions=(n_boxes, features_dims...).

mozuma.predictions.BatchVideoFramesPrediction dataclass

Result of video frames extraction on a video

Attributes:

Name	Type	Description
frame_indices	Sequence[int]	The frame index in the source video
features	ArrayLike | None	The features of the video frames. Dimensions=(n_frames, features_dims...)