vis4d.model.track3d.cc_3dt

CC-3DT model implementation.

This file composes the operations associated with CC-3DT `https://arxiv.org/abs/2212.01247`_ into the full model implementation.

Classes

CC3DT([basemodel, qdtrack_head, ...])

CC-3DT with custom detection results.

FasterRCNNCC3DT(num_classes[, basemodel, ...])

CC-3DT with Faster-RCNN detector.

FasterRCNNCC3DTOut(detector_3d_out, ...)

Output of CC-3DT model with Faster R-CNN detector.
class CC3DT(basemodel=None, qdtrack_head=None, track_graph=None, detection_range=None)[source]

CC-3DT with custom detection results.

Creates an instance of the class.

Parameters:

  • basemodel (BaseModel, optional) – Base model network. Defaults to None. If None, will use ResNet50.

  • qdtrack_head (QDTrack, optional) – QDTrack head. Defaults to None. If None, will use default QDTrackHead.

  • track_graph (CC3DTrackGraph, optional) – Track graph. Defaults to None. If None, will use default CC3DTrackGraph.

  • detection_range (Sequence[float], optional) – Detection range for each class. Defaults to None.

__call__(images_list, images_hw, intrinsics_list, extrinsics_list, frame_ids, pred_boxes3d, pred_boxes3d_classes, pred_boxes3d_scores, pred_boxes3d_velocities)[source]

Type definition for call implementation.

Return type:

Track3DOut

forward(images_list, images_hw, intrinsics_list, extrinsics_list, frame_ids, pred_boxes3d, pred_boxes3d_classes, pred_boxes3d_scores, pred_boxes3d_velocities)[source]

Forward inference stage.

Curretnly only work with single batch per gpu.

Return type:

Track3DOut

class FasterRCNNCC3DT(num_classes, basemodel=None, faster_rcnn_head=None, rcnn_box_decoder=None, qdtrack_head=None, track_graph=None, pure_det=False)[source]

CC-3DT with Faster-RCNN 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 (CC3DTrackGraph, optional) – Track graph. Defaults to None. If None, will use default CC3DTrackGraph.

  • pure_det (bool, optional) – Whether to use pure detection. Defaults to False.

__call__(images, images_hw, intrinsics, extrinsics=None, frame_ids=None, boxes2d=None, boxes3d=None, boxes3d_classes=None, boxes3d_track_ids=None, keyframes=None)[source]

Type definition for call implementation.

Return type:

FasterRCNNCC3DTOut | Track3DOut

forward(images, images_hw, intrinsics, extrinsics=None, frame_ids=None, boxes2d=None, boxes3d=None, boxes3d_classes=None, boxes3d_track_ids=None, keyframes=None)[source]

Forward.

Return type:

FasterRCNNCC3DTOut | Track3DOut

class FasterRCNNCC3DTOut(detector_3d_out: Tensor, detector_3d_target: Tensor, detector_3d_labels: Tensor, qdtrack_out: FasterRCNNQDTrackOut)[source]

Output of CC-3DT model with Faster R-CNN detector.

Create new instance of FasterRCNNCC3DTOut(detector_3d_out, detector_3d_target, detector_3d_labels, qdtrack_out)

detector_3d_labels: Tensor

Alias for field number 2

detector_3d_out: Tensor

Alias for field number 0

detector_3d_target: Tensor

Alias for field number 1

qdtrack_out: FasterRCNNQDTrackOut

Alias for field number 3