Event
DepTR-MOT: Depth-Refined Multi-Object Tracking
1 reports · 1 sources
Buyin Deng and co-authors propose DepTR-MOT, a DETR-based class-agnostic detector that embeds instance-level depth perception into the detector through instance-level soft depth label supervision and dense depth map distillation, and outputs instance depth at inference time to improve multi-object tracking under occlusion and close-range interactions. The authors state that the method strictly follows the mainstream tracking-by-detection (TBD) paradigm and requires no additional training or inference stages; the source code will be made publicly available.
Report timeline
- DepTR-MOT: Depth-Informed Trajectory Refinement for Multi-Object Tracking
DepTR-MOT proposes a category-agnostic DETR-based detector that embeds instance-level depth awareness into the detector through instance-level soft depth label supervision and dense depth map distillation, and outputs instance depth at inference to improve multi-object tracking under occlusion and close-proximity interactions.
arXiv cs.RO ·