Saved in:
Bibliographic Details
Main Authors: Sun, Yuyang, Wu, Yongliang, Zhu, Xingyu, Chen, Yuxia, Jiang, Zhenxiang, Ji, Yangguang, Zhu, Wenbo, Shi, Yanxi, Wu, Jay, Wang, Shuo, Yang, Xu
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2606.01104
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916070938378240
author Sun, Yuyang
Wu, Yongliang
Zhu, Xingyu
Chen, Yuxia
Jiang, Zhenxiang
Ji, Yangguang
Zhu, Wenbo
Shi, Yanxi
Wu, Jay
Wang, Shuo
Yang, Xu
author_facet Sun, Yuyang
Wu, Yongliang
Zhu, Xingyu
Chen, Yuxia
Jiang, Zhenxiang
Ji, Yangguang
Zhu, Wenbo
Shi, Yanxi
Wu, Jay
Wang, Shuo
Yang, Xu
contents VRR-QA evaluates whether video-language systems can infer spatial, temporal, viewpoint, depth, and visibility relations that are not always resolved by a single frame. We present an inference-only system built around adaptive test-time computation. The system first answers each question with a direct video-language model pass, then uses multiple lightweight views to find unstable questions. Only these difficult questions are routed to a high-budget dense evidence module that constructs timestamped frame observations, relation-specific probes, candidate verification, and conservative temporal aggregation. This design separates two problems that are often confused in video question answering: finding plausible alternative answers and deciding when a current answer should actually be changed. On the test split, the final system obtains 90.07 average accuracy and 87.81 macro average accuracy. The report focuses on the final test system and the implementation settings required to reproduce the adaptive dense verifier.
format Preprint
id arxiv_https___arxiv_org_abs_2606_01104
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Adaptive Dense Evidence Refinement for Video Relational Reasoning for VRR-QA Challenge
Sun, Yuyang
Wu, Yongliang
Zhu, Xingyu
Chen, Yuxia
Jiang, Zhenxiang
Ji, Yangguang
Zhu, Wenbo
Shi, Yanxi
Wu, Jay
Wang, Shuo
Yang, Xu
Computer Vision and Pattern Recognition
VRR-QA evaluates whether video-language systems can infer spatial, temporal, viewpoint, depth, and visibility relations that are not always resolved by a single frame. We present an inference-only system built around adaptive test-time computation. The system first answers each question with a direct video-language model pass, then uses multiple lightweight views to find unstable questions. Only these difficult questions are routed to a high-budget dense evidence module that constructs timestamped frame observations, relation-specific probes, candidate verification, and conservative temporal aggregation. This design separates two problems that are often confused in video question answering: finding plausible alternative answers and deciding when a current answer should actually be changed. On the test split, the final system obtains 90.07 average accuracy and 87.81 macro average accuracy. The report focuses on the final test system and the implementation settings required to reproduce the adaptive dense verifier.
title Adaptive Dense Evidence Refinement for Video Relational Reasoning for VRR-QA Challenge
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2606.01104