def __getitem__()

in pytorchvideo/data/domsev.py [0:0]

• 33 lines of code
• 9 McCabe index (conditional complexity)

    def __getitem__(self, index) -> Dict[str, Any]:
        """
        Samples a video clip associated to the given index.

        Args:
            index (int): index for the video clip.

        Returns:
            A video clip with the following format if transform is None.

            .. code-block:: text

                {{
                    'video_id': <str>,
                    'video': <video_tensor>,
                    'audio': <audio_tensor>,
                    'labels': <labels_tensor>,
                    'start_time': <float>,
                    'stop_time': <float>
                }}
        """
        clip = self._clips[index]

        # Filter labels by only the ones that appear within the clip boundaries,
        # and unpack the labels so there is one per frame in the clip
        labels_in_video = self._labels_per_video[clip.video_id]
        labels_in_clip = []
        for label_data in labels_in_video:
            overlap_period = _get_overlap_for_time_range_pair(
                clip.start_time,
                clip.stop_time,
                label_data.start_time,
                label_data.stop_time,
            )
            if overlap_period is not None:
                overlap_start_time, overlap_stop_time = overlap_period

                # Convert the overlapping period between clip and label to
                # 0-indexed start and stop frame indexes, so we can unpack 1
                # label per frame.
                overlap_start_frame = _seconds_to_frame_index(
                    overlap_start_time, self._frames_per_second
                )
                overlap_stop_frame = _seconds_to_frame_index(
                    overlap_stop_time, self._frames_per_second
                )

                # Append 1 label per frame
                for _ in range(overlap_start_frame, overlap_stop_frame):
                    labels_in_clip.append(label_data)

        # Convert the list of LabelData objects to a tensor of just the label IDs
        label_ids = [labels_in_clip[i].label_id for i in range(len(labels_in_clip))]
        label_ids_tensor = torch.tensor(label_ids)

        clip_data = {
            "video_id": clip.video_id,
            **self._videos[clip.video_id].get_clip(clip.start_time, clip.stop_time),
            "labels": label_ids_tensor,
            "start_time": clip.start_time,
            "stop_time": clip.stop_time,
        }

        if self._transform:
            clip_data = self._transform(clip_data)

        return clip_data