qhjqhj00 got10k-eval
ReasoningTrack: Chain-of-Thought Reasoning for Long-term Vision-Language Tracking — Xiao Wang et al. (2025) (arXiv:2508.05221, 2025)
What this evaluates
Evaluates tracking performance on a large-scale dataset of 10,000 videos with diverse object categories, testing generalization to real-world scenarios with language guidance.
Datasets
- GOT-10k — total 10000; splits: test (-1)
Metrics
AO(primary) — range: percent- Average Overlap: mean IoU between predicted and ground-truth boxes across all frames.
SR0.5— range: percent- Success Rate at IoU threshold 0.5: percentage of frames with IoU > 0.5.
SR0.75— range: percent- Success Rate at IoU threshold 0.75: percentage of frames with IoU > 0.75.
Input / output format
Input: Video frames, initial language description, and ground-truth bounding boxes.
Output: Predicted bounding box coordinates per frame.
Scoring recipe
For each frame, compute IoU(pred, GT).
AO = mean(IoU)
SR0.5 = (count(IoU > 0.5) / N) * 100
SR0.75 = (count(IoU > 0.75) / N) * 100
Common pitfalls
- GOT-10k test set lacks language annotations; authors use external annotations from Guo's work.
- Large scale requires efficient inference; language update overhead must be managed.
Evidence (verbatim from paper)
The experimental results presented in Table[IV] indicate that after introducing the strategy of COT text update mechanism, our method achieves competitive results (AO:77.8%, SR0.5:88.5, SR0.75:77.0) on this dataset, significantly enhancing the performance of the baseline tracker.
Citation
@misc{wang2025reasoningtrack,
title={ReasoningTrack: Chain-of-Thought Reasoning for Long-term Vision-Language Tracking},
author={Xiao Wang et al. (2025)},
year={2025},
note={arXiv:2508.05221}
}
- arXiv: 2508.05221