Research map / Video / Video-language understanding
UniComp: Rethinking Video Compression Through Informational Uniqueness
Chao Yuan, Shimin Chen, Minliang Lin, Limeng Qiao, Guanglu Wan, Lin Ma
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 frame group fusion with fused frames instead of keeping first frame (ablation) | UniComp (32f, 20%) · LongVideoBench | accuracy: 56.9 -> 57.7 | Improved |
| add adaptive token allocation instead of uniform (ablation) | UniComp (32f, 20%) · MLVU | accuracy: 61.4 -> 64.4 | Improved |
| use SDC with uniqueness instead of attention top-k (ablation) | UniComp (32f, 20%) · VideoMME | accuracy: 57.4 -> 58.7 | Improved |
| add neighbor token fusion (ablation) | UniComp SDC · VideoMME | accuracy: 58.0 -> 58.7 | Improved |
| keep tokens in original id order instead of uniqueness order (ablation) | UniComp SDC · LongVideoBench | accuracy: 57.0 -> 57.7 | Mixed |
| add FGF and SDC+TA to 320-frame input (ablation) | LLaVA-OneVision-7B · VideoMME | accuracy: 56.5 -> 60.7 | Improved |
| compress to 10% at 320 frames vs uncompressed | UniComp · TTFT | speedup: 4.15x | Improved |
| use UniComp at 25% retention vs vanilla | LLaVA-OneVision-7B · avg accuracy | accuracy: 59.95 -> 60.78 | Improved |
| use UniComp at 10% retention vs vanilla | LLaVA-OneVision-7B · avg accuracy | accuracy: 59.95 -> 59.80 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports UniComp as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- EchoPrune: Interpreting Redundancy as Temporal Echoes for Efficient VideoLLMs (NeurIPS 2026 · better on 2 of 3 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- HoliTom — HoliTom: Holistic Token Merging for Fast Video Large Language Models (NeurIPS 2025) · better on 7 of 7 shared results
- FastVid — FastVID: Dynamic Density Pruning for Fast Video Large Language Models (NeurIPS 2025) · better on 3 of 3 shared results
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
- Towards Unified Human Perception and Machine Understanding: Token Flow Guided Compression Framework (CVPR 2026 · similarity 0.88)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.88)
- Prune Redundancy, Preserve Essence: Vision Token Compression in VLMs via Synergistic Importance-Diversity (ICLR 2026 · similarity 0.87)
- PVC: Progressive Visual Token Compression for Unified Image and Video Processing in Large Vision-Language Models (CVPR 2025 · similarity 0.87)
- UniCompress: Token Compression for Unified Vision-Language Understanding and Generation (CVPR 2026 · similarity 0.86)
- ApET: Approximation-Error Guided Token Compression for Efficient VLMs (CVPR 2026 · similarity 0.86)
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.
