Research map / Video / Video-language understanding
PerceptionComp: A Video Benchmark for Complex Perception-Centric Reasoning
Shaoxuan Li, Zhixuan Zhao, Hanze Deng, Zirun Ma, Shulin Tian, Zuyan Liu, Yushi Hu, Haoning Wu, Yuhao Dong, Benlin Liu, Ziwei Liu, Ranjay Krishna
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 |
|---|---|---|---|
| use thinking model GPT-o3 instead of GPT-4o | GPT-4o · PerceptionComp | accuracy: +11.04% | Improved |
| use Gemini-2.5-Pro instead of Gemini-2.5-Flash | Gemini-2.5-Flash · PerceptionComp | accuracy: +6.19% | Improved |
| use thinking variant of Qwen3-VL-8B | Qwen3-VL-8B · PerceptionComp | accuracy: 34.06 -> 33.82 | No clear effect |
| use reasoning-trained VideoChat-R1 7B instead of Qwen2.5-VL 7B | Qwen2.5-VL-7B · PerceptionComp | accuracy: 22.73 -> 28.63 | Improved |
| use reasoning-trained Video-R1 7B instead of Qwen2.5-VL 7B | Qwen2.5-VL-7B · PerceptionComp | accuracy: 22.73 -> 26.27 | Improved |
| increase input frames from 16 to 50 (ablation) | GPT-o3 · PerceptionComp (100-video subset) | accuracy: 34.0 -> 43.54 | Improved |
| increase input frames from 16 to 64 (ablation) | Qwen3-VL-8B · PerceptionComp (100-video subset) | accuracy: 27.0 -> 34.80 | Improved |
| increase thinking-token budget from 1,024 to 8,192 (ablation) | Gemini-2.5-Flash · PerceptionComp (100-video subset) | accuracy | Improved |
| scale Qwen3-VL from 8B to 235B | Qwen3-VL · PerceptionComp | accuracy: 34.06 -> 34.02 | No clear effect |
| scale Qwen2.5-VL from 7B to 72B | Qwen2.5-VL · PerceptionComp | accuracy: 22.73 -> 31.33 | Improved |
| scale Qwen3-VL-Thinking from 8B to 235B | Qwen3-VL-Thinking · PerceptionComp | accuracy: 33.82 -> 38.20 | Improved |
| allow unrestricted rewatching for humans (ablation) | Human · PerceptionComp | accuracy: 18.97 -> 85.10 | Improved |
Closest papers by meaning
- VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding (NeurIPS 2025 · similarity 0.91)
- Act2See: Emergent Active Visual Perception for Video Reasoning (CVPR 2026 · similarity 0.91)
- MVVBench: Benchmarking 4D Reasoning in Vision-Language Models (NeurIPS 2026 · similarity 0.90)
- Video-of-Thought: Step-by-Step Video Reasoning from Perception to Cognition (ICML 2024 · similarity 0.90)
- Unleashing Perception-Time Scaling to Multimodal Reasoning Models (ICLR 2026 · similarity 0.90)
- BLINK-Twice: You see, but do you observe? A Reasoning Benchmark on Visual Perception (NeurIPS 2025 · similarity 0.90)
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.
