Research map / Video / Video-language understanding
Chain-of-Glimpse: Search-Guided Progressive Object-Grounded Reasoning for Video Understanding
Zhixuan Wu, Quanxing Zha, Teng Wang, Genbao Xu, Wenyuan Gu, Wei Rao, Nan Ma, Bo Cheng, Soujanya Poria
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add object-grounded reasoning with MCTS, SFT and GRPO (full method) to Qwen2.5-VL-7B | Qwen2.5-VL-7B · NExTQA | accuracy (%): 79.7 -> 83.3 | Improved |
| add object-grounded reasoning with random object selection (ablation) | Qwen2.5-VL-7B · NExTQA | accuracy (%): 79.7 -> 79.7 | No clear effect |
| use greedy object exploration instead of random (ablation) | Qwen2.5-VL-7B · NExTQA | accuracy (%): 79.7 -> 80.9 | Improved |
| add plain CoT reasoning (ablation) | Qwen2.5-VL-7B · Video-Holmes | accuracy (%): 38.8 -> 38.6 | No clear effect |
| add MCTS-distilled SFT (ablation) | Qwen2.5-VL-7B · NExTQA | accuracy (%): 79.7 -> 80.6 | Improved |
| add GRPO after SFT without grounding (ablation) | SFT model · NExTQA | accuracy (%): 80.6 -> 81.4 | Improved |
| add MCTS-based OGVRF to SFT (ablation) | SFT model · NExTQA | accuracy (%): 80.6 -> 82.3 | Improved |
| use evidence-aware reward instead of vanilla GRPO on OGVRF (ablation) | Chain-of-Glimpse · NExTQA | accuracy (%): 82.7 -> 83.3 | Improved |
| use evidence-aware reward instead of vanilla GRPO on OGVRF (ablation) | Chain-of-Glimpse · Video-Holmes | accuracy (%): 41.8 -> 42.5 | Improved |
| add OGVRF for long videos | Qwen2.5-VL-7B · NExTQA Long videos | accuracy (%): 77.9 -> 83.2 | Improved |
| increase MCTS rollouts from 2 to 8 (ablation) | Chain-of-Glimpse-7B · NExT-GQA | Acc@GQA: 30.5 -> 32.1 | Improved |
| increase MCTS rollouts from 8 to 10 (ablation) | Chain-of-Glimpse-7B · NExT-GQA | Acc@GQA: 32.1 -> 30.9 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VideoAgent — VideoAgent: A Memory-augmented Multimodal Agent for Video Understanding (ECCV 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- When Thinking Drifts: Evidential Grounding for Robust Video Reasoning (NeurIPS 2025 · similarity 0.91)
- Video-of-Thought: Step-by-Step Video Reasoning from Perception to Cognition (ICML 2024 · similarity 0.90)
- Motion-Grounded Video Reasoning: Understanding and Perceiving Motion at Pixel Level (CVPR 2025 · similarity 0.89)
- Look, Remember and Reason: Grounded Reasoning in Videos with Language Models (ICLR 2024 · similarity 0.89)
- VideoReasonBench: Can MLLMs Perform Vision-Centric Complex Video Reasoning? (ICLR 2026 · similarity 0.89)
- Act2See: Emergent Active Visual Perception for Video Reasoning (CVPR 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
