LiftTrack: Source Code and RGB-T Epoch-60 Weights

LiftTrack source code, experiment configurations, dataset split lists, and training and evaluation scripts for RGB-T, RGB-D, and RGB-E visual object tracking, together with RGB-T inference weights from training epoch 60 and a checksum file.

The RGB-T model tensors are identical to the original training checkpoint and have passed strict loading and a CUDA forward comparison. Their correspondence with the final paper results has not yet been verified. The weights are for inference and omit optimizer, scheduler, RNG state, and local settings. RGB-D and RGB-E weights, benchmark datasets, and external pretrained model files are not included.

Corresponding author: Mengjie Hu. Source code and third-party notices are available at https://github.com/Star-Swift/LiftTrack. The original RGB-T weight release is https://github.com/Star-Swift/LiftTrack/releases/tag/rgbt-epoch60.