RGBT Tracking via All-layer Multimodal Interactions with Progressive Fusion Mamba

We use the LasHeR training set for training, RGBT210, RGBT234, LasHeR testing set, VTUAVST for testing, and their project addresses are as follows:

• RGBT210
• RGBT234
• LasHeR
• VTUAV

1. Our code is trained and tested with Python == 3.8.13, PyTorch == 2.1.1 and CUDA == 11.8 on NVIDIA GeForce RTX 4090.
2. Install causal_conv1d and mamba from our repo.

1. We adopt OSTrack as our base tracker. For our best result, we need to load the parameter from DropMAE.
2. Modify the relevant dataset and pretrained model paths, Then run the following command.

python lib/train/run_training.py --script ostrack_twobranch --config 384 --save_dir your_save_dir

Modify the relevant dataset and checkpoint paths, then run the following command.

python tracking/test.py --tracker_name ostrack_twobranch --tracker_param 384 --checkpoint_path your_checkpoint_path