UATrack

Pytorch implementation of the paper "Uncertainty-aware RGBT Tracking".

Introduction

• UATrack leverages uncertainty to enhance robust representation learning and reliable multimodal fusion for RGBT tracking, effectively addressing the challenges arising from dynamically varying modality quality in complex scenarios.
• UATrack achieves SOTA result on GTOT, RGBT210, RGBT234 and LasHeR datasets.

Uncertainty Scores

Note: This is an extended version of our conference paper TUMFNet published at IJCAI 2025.

Models and Results

You can download the model and results from here [Extraction Code: RGBT].

Path Setting

Run the following command to set paths:

cd <PATH>
python tracking/create_default_local_file.py --workspace_dir . --data_dir <PATH_of_Datasets> --save_dir ./output

You can also modify paths by these two files:

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

Training

Dowmload the pretrained foundation model (OSTrack and DropMae) and put it under ./pretrained/.

CUDA_VISIBLE_DEVICES=0,1  NCCL_P2P_LEVEL=NVL nohup  python tracking/train.py --script drgbt --config DRGBT603 --save_dir ./output --mode multiple --nproc_per_node 2 >  train_track.log &

Test

bash eval_rgbt.sh

Evaluation for LasHeR, RGBT234, RGBT210 and GTOT

python eval_lasher.py
python eval_rgbt234.py
python eval_rgbt210.py
python eval_gtot.py

Results

Acknowledgment

• This repo is based on BAT which is an exellent work, helps us to quickly implement our ideas.
• Thanks for the OSTrack and PyTracking library.