Awesome

DropTrack

The official fine-tuning implementation of DropTrack for the CVPR 2023 paper DropMAE: Masked Autoencoders with Spatial-Attention Dropout for Tracking Tasks.

:sunny: Highlights

* Thanks for the great OSTrack library, which helps us to quickly implement the DropMAE VOT fine-tuning. The repository mainly follows the OSTrack repository.

* The OSTrack w/ our DropMAE pre-trained models can achieve state-of-the-art performance on existing popular tracking benchmarks.

:star2: Training Speed

Our DropTrack has the same training procedure and nearly the same model parameters (i.e., except for using two frame identity embeddings) w/ OSTrack, so the training speed is consistent w/ OSTrack. We use 4 A100 GPUs w/ a total batch size of 128, which costs about ~6 hours (100 Epochs) for training on GOT-10k.

Install the environment

Option1: The Anaconda is used to create the Python environment, which mainly follows the installation in OSTrack. The specific installation packages are listed in requirements.txt for consideration, which can be installed in the following way:

conda create -n droptrack python=3.8
conda activate droptrack
pip install -r requirements.txt

Set project paths

Run the following command to set paths for this project

python tracking/create_default_local_file.py --workspace_dir . --data_dir ./data --save_dir ./output

After running this command, you can also modify paths by editing these two files

lib/train/admin/local.py  # paths about training
lib/test/evaluation/local.py  # paths about testing

Data Preparation

Put the tracking datasets in ./data. It should look like:

${PROJECT_ROOT}
 -- data
     -- lasot
         |-- airplane
         |-- basketball
         |-- bear
         ...
     -- got10k
         |-- test
         |-- train
         |-- val
     -- coco
         |-- annotations
         |-- images
     -- trackingnet
         |-- TRAIN_0
         |-- TRAIN_1
         ...
         |-- TRAIN_11
         |-- TEST

Training

• Download pre-trained DropMAE models in DropMAE and put it under $PROJECT_ROOT$/pretrained_models.
• Modify the PRETRAIN_FILE tag in vitb_384_mae_ce_32x4_ep300.yaml or vitb_384_mae_ce_32x4_got10k_ep100.yaml to the name of your downloaded DropMAE pre-trained models.
• Training Command on GOT-10K:

cd path_to_your_project
python tracking/train.py --script ostrack --config vitb_384_mae_ce_32x4_got10k_ep100 --save_dir sabe_path --mode multiple --nproc_per_node 4 --use_lmdb 0 --use_wandb 0

• Training Command on the other tracking datasets:

cd path_to_your_project
python tracking/train.py --script ostrack --config vitb_384_mae_ce_32x4_ep300 --save_dir save_path --mode multiple --nproc_per_node 4 --use_lmdb 0 --use_wandb 0

Training logs

The training log of DropTrack-Got10k-100E is available here.

Evaluation

Download the tracking model weights

Change the corresponding values of lib/test/evaluation/local.py to the actual benchmark saving paths. Note that the save_dir tag should be set to the downloaded tracking model path and you can also modify the tracking model name in lib/test/parameter/ostrack.py.

Some testing examples:

• LaSOT or other off-line evaluated benchmarks (modify --dataset correspondingly)

python tracking/test.py ostrack vitb_384_mae_ce_32x4_ep300 --dataset lasot --threads 16 --num_gpus 4
python tracking/analysis_results.py # need to modify tracker configs and names

• GOT10K-test

python tracking/test.py ostrack vitb_384_mae_ce_32x4_got10k_ep100 --dataset got10k_test --threads 16 --num_gpus 4
python lib/test/utils/transform_got10k.py --tracker_name ostrack --cfg_name vitb_384_mae_ce_32x4_got10k_ep100

Acknowledgments

• Thanks for the OSTrack library for convenient implementation.

Citation

If our work is useful for your research, please consider cite:

@inproceedings{dropmae2023,
  title={DropMAE: Masked Autoencoders with Spatial-Attention Dropout for Tracking Tasks},
  author={Qiangqiang Wu and Tianyu Yang and Ziquan Liu and Baoyuan Wu and Ying Shan and Antoni B. Chan},
  booktitle={CVPR},
  year={2023}
}