fzh0917 / SparseTT

The official implementation for paper "SparseTT: Visual Tracking with Sparse Transformers"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

SparseTT

The official implementation for paper "SparseTT: Visual Tracking with Sparse Transformers".

This paper is accepted by IJCAI2022 as a long oral presentation.

Installation

  • Prepare Anaconda, CUDA and the corresponding toolkits. CUDA version required: 11.3.

  • Create a new conda environment and activate it.

conda create -n SparseTT python=3.7 -y
conda activate SparseTT
  • Install pytorch and torchvision.
conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cudatoolkit=11.3 -c pytorch -c conda-forge
# pytorch version: >= 1.9.0 
  • Install other required packages.
pip install -r requirements.txt

Test

  • Prepare the datasets: OTB2015, VOT2018, UAV123, GOT-10k, TrackingNet, LaSOT, ILSVRC VID*, ILSVRC DET*, COCO*, and something else you want to test. Set the paths as the following:
├── SparseTT
|   ├── ...
|   ├── ...
|   ├── datasets
|   |   ├── COCO -> /opt/data/COCO
|   |   ├── GOT-10k -> /opt/data/GOT-10k
|   |   ├── ILSVRC2015 -> /opt/data/ILSVRC2015
|   |   ├── LaSOT -> /opt/data/LaSOT/LaSOTBenchmark
|   |   ├── OTB
|   |   |   └── OTB2015 -> /opt/data/OTB2015
|   |   ├── TrackingNet -> /opt/data/TrackingNet
|   |   ├── UAV123 -> /opt/data/UAV123/UAV123
  • Notes

i. Star notation(*): just for training. You can ignore these datasets if you just want to test the tracker.

ii. In this case, we create soft links for every dataset. The real storage location of all datasets is /opt/data/. You can change them according to your situation.

  • Download the pretrained models.

    📎 GOT-10k model 📎 fulldata model

  • Set the pretrained model path for the item pretrain_model_path in the configuration file, then run shell commands.

  • Note that all paths we used here are relative, not absolute. See any configuration file in the experiments directory for examples and details.

GOT-10k

python main/test.py --config experiments/sparsett/test/got10k/sparsett_swin_got10k.yaml

LaSOT

python main/test.py --config experiments/sparsett/test/lasot/sparsett_swin_lasot.yaml

TrackingNet

python main/test.py --config experiments/sparsett/test/trackingnet/sparsett_swin_trackingnet.yaml

UAV123

python main/test.py --config experiments/sparsett/test/uav123/sparsett_swin_uav123.yaml

OTB2015

python main/test.py --config experiments/sparsett/test/otb2015/sparsett_swin_otb2015.yaml

Training

  • Prepare the datasets as described in the last subsection.
  • Download the pretrained backbone model from here, and put it in the SparseTT/models/swin/ directory.
  • Run the shell command.

GOT-10k

python main/train.py --config experiments/sparsett/train/got10k/sparsett_swin_train_got10k.yaml

fulldata

python main/train.py --config experiments/sparsett/train/fulldata/sparsett_swin_train_fulldata.yaml

Testing Results

Click here to download all testing results that includes:

  • LaSOT
  • TrackingNet
  • GOT-10k
  • UAV123
  • OTB2015

Acknowledgement

Repository

This repository is built on the top of the single object tracking framework video_analyst. See it for more instructions and details.

References

@article{fu2022sparsett,
  title={SparseTT: Visual Tracking with Sparse Transformers},
  author={Fu, Zhihong and Fu, Zehua and Liu, Qingjie and Cai, Wenrui and Wang, Yunhong},
  booktitle={IJCAI},
  year={2022}
}

Contact

If you have any questions, just create issues or email me:smile:.

About

The official implementation for paper "SparseTT: Visual Tracking with Sparse Transformers"

License:MIT License


Languages

Language:Python 98.1%Language:Cython 1.0%Language:C 0.9%Language:Shell 0.0%