TransT-M - High-performance Transformer Tracking

Official implementation of the TransT-M, including training code and trained models. Models

Installation

This document contains detailed instructions for installing the necessary dependencied for TransT-M. The instructions have been tested on Ubuntu 18.04 system.

Install dependencies

• Create and activate a conda environment

conda create -n transt python=3.7
conda activate transt

• Install PyTorch

conda install -c pytorch pytorch=1.5 torchvision

• Install other packages

conda install matplotlib pandas tqdm
pip install opencv-python tb-nightly visdom scikit-image tikzplotlib gdown
conda install cython scipy
pip install pycocotools jpeg4py
pip install wget
pip install shapely==1.6.4.post2

• Setup the environment
  Create the default environment setting files.

# Change directory to <PATH_of_TransT>
cd TransT-M

# Environment settings for pytracking. Saved at pytracking/evaluation/local.py
python -c "from pytracking.evaluation.environment import create_default_local_file; create_default_local_file()"

# Environment settings for ltr. Saved at ltr/admin/local.py
python -c "from ltr.admin.environment import create_default_local_file; create_default_local_file()"

You can modify these files to set the paths to datasets, results paths etc.

• Add the project path to environment variables
  Open ~/.bashrc, and add the following line to the end. Note to change <path_of_TransT> to your real path.

export PYTHONPATH=<path_of_TransT>:$PYTHONPATH

• Download the pre-trained networks Download the network for TransT-M and put it in the directory set by "network_path" in "pytracking/evaluation/local.py". By default, it is set to pytracking/networks.

Quick Start

TRAINING

• Modify local.py to set the paths to datasets, results paths etc.
• Runing the following commands to train the TransT-M. You can customize some parameters by modifying the settings in transt

1. Train the base model of TransT-M

conda activate transt
cd TransT-M/ltr
python -m torch.distributed.launch --nproc_per_node 8 run_training_multigpu.py transt transt

2. Train the iou head of TransT-M, you should set a new workspace_dir in local.py and modify the settings.transt_path in transt_iou.py to the path of a trained base transt model

python -m torch.distributed.launch --nproc_per_node 8 run_training_multigpu.py transt transt_iou

3. Train the segmentation branch of TransT-M, you should set a new workspace_dir in local.py and modify the settings.transt_path in transt_iou_seg.py to the path of a trained transt_iou model

python -m torch.distributed.launch --nproc_per_node 8 run_training_multigpu.py transt transt_iou_seg

Evaluation

• We integrated PySOT for evaluation You need to specify the path of the model and dataset in the following files: test_got.py, test_lasot.py, test_nfs.py, test_otb.py, test_tracking.py, test_uav.py

  net_path = '/path_to_model' #Absolute path of the model
  dataset_root= '/path_to_datasets' #Absolute path of the datasets

  You need to specify the path of dataset in eval.py

  root = '/path_to_datasets' #Absolute path of the datasets

  Then run the following commands

  conda activate TransT
  cd TransT-M
  python -u pysot_toolkit/test_lasot.py --dataset LaSOT #test tracker
  python pysot_toolkit/eval.py --tracker_path pysot_toolkit/results/ --dataset LaSOT --num 1 #eval tracker
  
  python -u pysot_toolkit/test_got.py --dataset GOT-10k #test tracker
  
  python -u pysot_toolkit/test_trackingnet.py --dataset Tracking #test tracker
  
  python -u pysot_toolkit/test_nfs.py --dataset NFS #test tracker
  python pysot_toolkit/eval.py --tracker_path pysot_toolkit/results/ --dataset NFS --num 1 #eval tracker
  
  python -u pysot_toolkit/test_uav.py --dataset UAV #test tracker
  python pysot_toolkit/eval.py --tracker_path pysot_toolkit/results/ --dataset UAV --num 1 #eval tracker
  
  python -u pysot_toolkit/test_otb.py --dataset OTB #test tracker
  python pysot_toolkit/eval.py --tracker_path pysot_toolkit/results/ --dataset OTB --num 1 #eval tracker

• For evaluation on VOT2021, run the following commands. You should modify the paths in trackers.ini, and the net path in transt_VOT2021.py

  cd TransT-M/vot2021_workspace
  vot evaluate TransT_M

Acknowledgement

This is a modified version of the python framework PyTracking based on Pytorch, also borrowing from PySOT and GOT-10k Python Toolkit. We would like to thank their authors for providing great frameworks and toolkits.

Contact

• Xin Chen (email:chenxin3131@mail.dlut.edu.cn)