Research map / Video / Video-language understanding
Video-BCI: Bayesian Cognitive Integration of Self-Prior Hypotheses for Video Understanding
Xing Xi, Peixian Chen, Yu Qiu, Ronghua Luo, Peilin Tong, Jiahao Liang
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 Dialectical Uncertainty Signal to GRPO (ablation) | Video-BCI (Qwen2.5-VL-7B) · VSI-Bench (16 frames) | accuracy: 32.0 -> 36.2 | Improved |
| add DUS to GRPO (ablation) | Video-BCI · MMVU (16 frames) | accuracy: 64.8 -> 68.0 | Improved |
| add Process Tracing Signal to GRPO (ablation) | Video-BCI · MMVU (16 frames) | accuracy: 64.8 -> 68.2 | Improved |
| add PTS to GRPO (ablation) | Video-BCI · TempCompass (16 frames) | accuracy: 72.3 -> 74.5 | Improved |
| add PTS on top of DUS (full Video-BCI-R) (ablation) | Video-BCI-7B · VideoMME (16 frames) | accuracy: 58.0 -> 58.7 | Improved |
| add PTS on top of DUS (ablation) | Video-BCI-7B · VSI-Bench (16 frames) | accuracy: 36.2 -> 35.7 | Hurt |
| add PTS on top of DUS (ablation) | Video-BCI-7B · MMVU (16 frames) | accuracy: 68.0 -> 67.8 | No clear effect |
| add PTS on top of DUS (ablation) | Video-BCI-7B · VideoMMMU (16 frames) | accuracy: 53.1 -> 52.4 | Hurt |
| add self-prior hypotheses as input (BCI without RL) vs CoT prompting (ablation) | Qwen2.5-VL-7B · MMVU (16 frames) | accuracy: 59.2 -> 66.3 | Improved |
| add DUS+PTS RL over majority voting (ablation) | Video-BCI-R-7B · VSI-Bench (16 frames) | accuracy: 28.5 -> 35.7 | Improved |
| train with BCI instead of GRPO | Video-BCI · training accuracy curve | training accuracy: plateau around 0.72-0.74 | Improved |
| increase frames from 16 to 64 | Video-BCI-R-7B · VideoMME | accuracy: 58.7 -> 62.9 | Improved |
Closest papers by meaning
- Video-of-Thought: Step-by-Step Video Reasoning from Perception to Cognition (ICML 2024 · similarity 0.91)
- When Thinking Drifts: Evidential Grounding for Robust Video Reasoning (NeurIPS 2025 · similarity 0.91)
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · similarity 0.89)
- SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models (ICML 2026 · similarity 0.89)
- Mimic Human Cognition, Master Multi-Image Reasoning: A Meta-Action Framework for Enhanced Visual Understanding (CVPR 2026 · similarity 0.89)
- Black Swan: Abductive and Defeasible Video Reasoning in Unpredictable Events (CVPR 2025 · similarity 0.89)
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.
