- Rating
- 1
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
BVTR replaces visual tokens according to each follow-up question while keeping the active token budget fixed. It updates the cache through dense reconstruction or sparse repair. SR1000 accuracy increases from 88.9% to 91.0%, and MT4 sequence success rises from 68.0% to 75.5%. Gains on VQAv2 and with strong initial selectors are limited.
Critical strengths
- Recovering visual evidence needed by later questions under a fixed token budget addresses a practical problem.
- Random and previous-question refresh controls, together with multi-turn and cross-model results, support the effectiveness of question-conditioned refresh.
Critical weaknesses
- Unclear benefit over simpler or stronger baselines. Global reselection achieves 90.9% versus 91.0% for BVTR. Refresh provides no gain with a strong initial selector. Ablations also show no advantage from selecting repair tokens by changes in cached keys and values.
- Incomplete efficiency evaluation. Sparse96 takes 2.06 times as long as Fixed. Full replays historical answers, although its unchanged visual context normally permits cache reuse. This choice needs justification. Incremental peak memory alone does not establish total memory requirements.
Questions for authors
- How are the datasets constructed and filtered? Does filtering depend on omissions or errors from a particular selector or model?
- With standard cache reuse for Full, what are each method’s latency and total peak GPU memory?
- On the full test set, how does BVTR compare with global reselection and updating only newly activated tokens in accuracy, latency, and memory?
- Please specify token scoring, replacement counts, and how historical question–answer caches are updated in turns three and four.