Skip to content

LPXTT/SimTrack

Repository files navigation

SimTrack

Official Code for Backbone is All Your Need: A Simplified Architecture for Visual Object Tracking accepted by ECCV 2022.

Requirements

  • python==3.8.15
  • torch==1.8.1
  • torchvision==0.9.0
  • timm==0.5.4

Results (AUC)

method FLOPs LaSOT TNL2K TrackingNet GOT-10k_Test UAV123 model-clip model-mae
SimTrack 25.0G 69.3 54.8 82.3 70.6 69.8 Sim-B/16 MAE-B/16
Raw Results - LaSOT TNL2K TrackingNet GOT-10k_Test UAV123 - -

It should be noted that the above pretrained model is trained under PYSOT codebase. If you want to load the model in current pipeline, you may need to change the parameter name. Or you can train a new model with the current pipeline.

Evaluation

Download the model Sim-B/16. Add the model path to

network.load_state_dict(torch.load('please add the model path here', map_location='cpu'), strict=True)

python tracking/test.py simtrack baseline --dataset got10k_test --threads 32

Before evaluation, please make sure the data path in local.py is correct.

Training

python tracking/train.py --script simtrack --config baseline_got10k_only --save_dir . --mode multiple --nproc_per_node 8

Before training, please make sure the data path in local.py is correct. Also, please make sure there is "ViT-B/16": "https://openaipublic.azureedge.net/clip/models/5806e77cd80f8b59890b7e101eabd078d9fb84e6937f9e85e4ecb61988df416f/ViT-B-16.pt" in '**/anaconda3/envs/stark/lib/python3.8/site-packages/clip/clip.py'. If not, please add it manually.

Thanks

This implementation is based on STARK. Please ref to their reposity for more details.

Citation

If you find that this project helps your research, please consider citing our paper:

@article{chen2022backbone,
  title={Backbone is All Your Need: A Simplified Architecture for Visual Object Tracking},
  author={Chen, Boyu and Li, Peixia and Bai, Lei and Qiao, Lei and Shen, Qiuhong and Li, Bo and Gan, Weihao and Wu, Wei and Ouyang, Wanli},
  journal={arXiv preprint arXiv:2203.05328},
  year={2022}
}

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published