This the official implementation of Looking Beyond Two Frames: End-to-End Multi-Object Tracking Using Spatial and Temporal Transformers. The code is built on top of MMTRACK and MMDET. Please refer to these two repo for setup. Paper Link:https://arxiv.org/abs/2103.14829
Install the main packages:
- torch==1.9.1
- mmcv==1.4.1
- mmdet==2.19.1
- mmtrack==0.8.0
Then use the following command to setup MO3TR:
bash setup
python run/train_nf.py
python run/track_nf.py
You can download the model weight used for MOT17 test here: https://drive.google.com/file/d/1d-8A2dRNyMLb4kCYa7DMHCDvuiSMxc5K/view?usp=sharing