vis4d.model.track.qdtrack¶

Quasi-dense instance similarity learning model.

Classes

`FasterRCNNQDTrack`(num_classes[, basemodel, ...])	Wrap QDTrack with Faster R-CNN detector.
`FasterRCNNQDTrackOut`(detector_out, ...)	Output of QDtrack model.
`YOLOXQDTrack`(num_classes[, basemodel, fpn, ...])	Wrap QDTrack with YOLOX detector.
`YOLOXQDTrackOut`(detector_out, key_images_hw, ...)	Output of QDtrack YOLOX model.

class FasterRCNNQDTrack(num_classes, basemodel=None, faster_rcnn_head=None, rcnn_box_decoder=None, qdtrack_head=None, track_graph=None, weights=None)[source]¶

Wrap QDTrack with Faster R-CNN detector.

Creates an instance of the class.

Parameters:

num_classes (int) – Number of object categories.
basemodel (BaseModel, optional) – Base model network. Defaults to None. If None, will use ResNet50.
faster_rcnn_head (FasterRCNNHead, optional) – Faster RCNN head. Defaults to None. if None, will use default FasterRCNNHead.
rcnn_box_decoder (DeltaXYWHBBoxDecoder, optional) – Decoder for RCNN bounding boxes. Defaults to None.
qdtrack_head (QDTrack, optional) – QDTrack head. Defaults to None. If None, will use default QDTrackHead.
track_graph (QDTrackGraph, optional) – Track graph. Defaults to None. If None, will use default QDTrackGraph.
weights (str, optional) – Weights to load for model.

__call__(images, images_hw, original_hw, frame_ids, boxes2d=None, boxes2d_classes=None, boxes2d_track_ids=None, keyframes=None)[source]¶

Type definition for call implementation.

Return type:: TrackOut | FasterRCNNQDTrackOut

forward(images, images_hw, original_hw, frame_ids, boxes2d=None, boxes2d_classes=None, boxes2d_track_ids=None, keyframes=None)[source]¶

Forward.

Return type:: TrackOut | FasterRCNNQDTrackOut

class FasterRCNNQDTrackOut(detector_out: FRCNNOut, key_images_hw: list[tuple[int, int]], key_target_boxes: list[Tensor], key_embeddings: list[Tensor], ref_embeddings: list[list[Tensor]], key_track_ids: list[Tensor], ref_track_ids: list[list[Tensor]])[source]¶

Output of QDtrack model.

Create new instance of FasterRCNNQDTrackOut(detector_out, key_images_hw, key_target_boxes, key_embeddings, ref_embeddings, key_track_ids, ref_track_ids)

detector_out: FRCNNOut¶: Alias for field number 0

key_embeddings: list[Tensor]¶: Alias for field number 3

key_images_hw: list[tuple[int, int]]¶: Alias for field number 1

key_target_boxes: list[Tensor]¶: Alias for field number 2

key_track_ids: list[Tensor]¶: Alias for field number 5

ref_embeddings: list[list[Tensor]]¶: Alias for field number 4

ref_track_ids: list[list[Tensor]]¶: Alias for field number 6

class YOLOXQDTrack(num_classes, basemodel=None, fpn=None, yolox_head=None, train_postprocessor=None, test_postprocessor=None, qdtrack_head=None, track_graph=None, weights=None)[source]¶

Wrap QDTrack with YOLOX detector.

Creates an instance of the class.

Parameters:

num_classes (int) – Number of object categories.
basemodel (BaseModel, optional) – Base model. Defaults to None. If None, will use CSPDarknet.
fpn (FeaturePyramidProcessing, optional) – Feature Pyramid Processing. Defaults to None. If None, will use YOLOXPAFPN.
yolox_head (YOLOXHead, optional) – YOLOX head. Defaults to None. If None, will use YOLOXHead.
train_postprocessor (YOLOXPostprocess, optional) – Post processor for training. Defaults to None. If None, will use YOLOXPostprocess.
test_postprocessor (YOLOXPostprocess, optional) – Post processor for testing. Defaults to None. If None, will use YOLOXPostprocess.
qdtrack_head (QDTrack, optional) – QDTrack head. Defaults to None. If None, will use default QDTrackHead.
track_graph (QDTrackGraph, optional) – Track graph. Defaults to None. If None, will use default QDTrackGraph.
weights (str, optional) – Weights to load for model.

__call__(images, images_hw, original_hw, frame_ids, boxes2d=None, boxes2d_classes=None, boxes2d_track_ids=None, keyframes=None)[source]¶

Type definition for call implementation.

Return type:: TrackOut | FasterRCNNQDTrackOut

forward(images, images_hw, original_hw, frame_ids, boxes2d=None, boxes2d_classes=None, boxes2d_track_ids=None, keyframes=None)[source]¶

Forward.

Return type:: TrackOut | YOLOXQDTrackOut

class YOLOXQDTrackOut(detector_out: YOLOXOut, key_images_hw: list[tuple[int, int]], key_target_boxes: list[Tensor], key_target_classes: list[Tensor], key_embeddings: list[Tensor], ref_embeddings: list[list[Tensor]], key_track_ids: list[Tensor], ref_track_ids: list[list[Tensor]])[source]¶

Output of QDtrack YOLOX model.

Create new instance of YOLOXQDTrackOut(detector_out, key_images_hw, key_target_boxes, key_target_classes, key_embeddings, ref_embeddings, key_track_ids, ref_track_ids)

detector_out: YOLOXOut¶: Alias for field number 0

key_embeddings: list[Tensor]¶: Alias for field number 4

key_images_hw: list[tuple[int, int]]¶: Alias for field number 1

key_target_boxes: list[Tensor]¶: Alias for field number 2

key_target_classes: list[Tensor]¶: Alias for field number 3

key_track_ids: list[Tensor]¶: Alias for field number 6

ref_embeddings: list[list[Tensor]]¶: Alias for field number 5

ref_track_ids: list[list[Tensor]]¶: Alias for field number 7