search

open-mmlab / mmtracking

OpenMMLab Video Perception Toolbox. It supports Video Object Detection (VID), Multiple Object Tracking (MOT), Single Object Tracking (SOT), Video Instance Segmentation (VIS) with a unified framework.
https://mmtracking.readthedocs.io/en/latest/
Apache License 2.0
3.55k stars 594 forks source link

Some questions about QDTrack #612

Open hcv1027 opened 2 years ago

hcv1027 commented 2 years ago

Hi, Thanks for your great work. It's my first time to use the open-mmlab project. I'm trying to implement a customized model based on QDTrack and train it with BDD100K dataset. I want to change the detector from two-stage FasterRCNN to one-stage detector like yolox.

I have some questions when I trace the code of QDTrack:

  1. When will the simple_test() be called? It looks like self.tracker only be used here.
  2. Dose self.tracker play a role during training stage?
  3. According to this reply, QuasiDenseTAOTracker can be used to train multi-classes dataset like bdd100k, does it correct?
JingweiZhang12 commented 2 years ago
  1. 'simple_test' is called in 'forward_test' in base.py.
  2. 'self.tracker' is only used in the inference, not used in the training stage.
  3. This reply is right, and you can refer to it.