Research map / Video / Video-language understanding
Small Vision-Language Models are Smart Compressors for Long Video Understanding
Junjie Fei, Jun Chen, Zechun Liu, Yunyang Xiong, Chong Zhou, Wei Wen, Junlin Han, Mingchen Zhuge, Saksham Suri, Qi Qian, Shuming Liu, Lemeng Wu and 4 more
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 Adaptive Token Allocation (8K) vs none (16K) (ablation) | Tempo · LVBench | accuracy: 51.1 -> 52.3 | Improved |
| add Stage 3 long-context SFT (ablation) | Tempo · LVBench | accuracy: 47.3 -> 51.1 | Improved |
| use ATA instead of uniform subsampling (ablation) | Tempo · LongVideoBench | accuracy: 61.9 -> 65.1 | Improved |
| route to lowest-scoring segments (adversarial) (ablation) | Tempo · LongVideoBench | accuracy: 65.1 -> 50.7 | Hurt |
| use ATA instead of hard top-K routing (ablation) | Tempo · LVBench | accuracy: 52.7 -> 52.3 | No clear effect |
| keep head tokens instead of tail tokens (dynamic truncation) (ablation) | Tempo · LongVideoBench | accuracy: 61.9 -> 65.1 | Improved |
| use head truncation instead of token merging (ablation) | Tempo · LVBench | accuracy: 53.0 -> 52.3 on LVBench; 66.3 -> 67.7 on Video-MME | Mixed |
| score relevance with Tempo SVLM and explicit prompt instead of external reranker (ablation) | Tempo · LVBench | accuracy: 51.8 -> 52.3 | No clear effect |
| use Tempo SVLM explicit prompt vs standard prompt for scoring (ablation) | Tempo · LVBench | accuracy: 53.4 -> 52.3 on LVBench; 64.1 -> 65.1 on LongVideoBench | Mixed |
| enforce minimal temporal anchors k_min=4 instead of hard pruning (ablation) | Tempo · LongVideoBench | accuracy: 63.9 -> 65.1 | Improved |
| raise budget from 4K to 8K | Tempo · LVBench | accuracy: 52.7 -> 52.3 | No clear effect |
| scale to 2048 frames with 12K budget | Tempo · LVBench | accuracy: 52.3 -> 53.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- VideoChat-Flash — VideoChat-Flash: Hierarchical Compression for Long-Context Video Modeling (ICLR 2026) · better on 5 of 5 shared results
- LongVU — LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding (ICML 2025) · better on 0 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
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.93)
- LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding (ICML 2025 · similarity 0.91)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.91)
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · similarity 0.90)
- ApET: Approximation-Error Guided Token Compression for Efficient VLMs (CVPR 2026 · similarity 0.90)
- Variation-aware Vision Token Dropping for Faster Large Vision-Language Models (CVPR 2026 · 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.
