Your TNL2k raw results is inflated?

OpenSpaceAI / UVLTrack

The official pytorch implementation of our AAAI 2024 paper "Unifying Visual and Vision-Language Tracking via Contrastive Learning"

MIT License

16 stars 2 forks source link

Your TNL2k raw results is inflated? #7

Closed laisimiao closed 3 months ago

laisimiao commented 3 months ago

It seems you use python analysis_results to evaluate tnl2k, which is higher than the official matlab toolkit. This is a common and wrong way to get the SUC on tnl2k in sot community.

568xiaoma commented 3 months ago

The evaluation and comparison of tnl2k performance are consistent with JointNLT (CVPR2023), which is our main baseline for vision-language tracking.