Awesome

Revisiting the RGBT Tracking Benchmarks from the Perspective of Modality Validity: A New Benchmark, Problem, and Solution

⭐ This main contributions are:

A new benchmark, MV-RGBT, and a new way for in-depth analysis of the methods.
A new problem, when to fuse, is introduced to explore the necessity of multi-modal fusion.
A new method, MoETrack, is derived with a mixture of experts.
New state-of-the-art results on several benchmarks, including MV-RGBT, RGBT234, LasHeR, and VTUAV-ST.
The code will be available here

🫵Find our survey work at another repo

Benchmark Data Comparison

⭐ MV-RGBT will be published after this work accepted!

<img src="ER_Cat_Lawn0.gif" width="800"> <img src="ET_Fish_River02.gif" width="800"> <img src="figs/data.png" width="600">

Selection Results

LasHeR:

<img src="figs/results-LasHeR.png" width="600">

MV-RGBT:

<img src="figs/results-MV-RGBT.png" width="600">