Research map / Video / Video-language understanding
Less Is More: Vision Representation Compression for Efficient Video Generation with Large Language Models
Yucheng Zhou, Jihai Zhang, Guanjie Chen, Jianbing Shen, Yu Cheng
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 representation compression (ratio 4) to Loong-style LLM video generator | Loong · Class-to-video FaceForensics | FVD: 46.11 -> 26.64 | Improved |
| add representation compression to Loong | Loong · Class-to-video SkyTimelapse | FVD: 62.71 -> 41.95 | Improved |
| add representation compression to Loong | Loong · Class-to-video UCF101 | FVD: 254.47 -> 250.53 | Improved |
| add representation compression to Loong | Loong · Class-to-video Taichi-HD | FVD: 105.53 -> 96.39 | Improved |
| compress tokens 4x instead of full sequence | Loong · 17-frame generation time, Base 100M | inference time (s): 12.32 -> 0.88 | Improved |
| compress tokens 4x instead of full sequence | Loong · 17-frame generation time, XLarge 770M | inference time (s): 26.08 -> 2.89 | Improved |
| raise compression rate from 4 to 16 or 64 (ablation) | VRC · Class-to-video datasets | FVD | Hurt |
| moderate compression (rate 4) versus no compression (ablation) | VRC · Class-to-video datasets | FVD | Improved |
| scale model from 100M to 770M (ablation) | VRC · Class-to-video datasets | FVD | Improved |
| use VRC vs Loong for memory | VRC · GPU memory | memory | Improved |
| VRC vs VideoPoet on zero-shot text-to-video | VRC · MSR-VTT | FVD: 213 -> 268 | Hurt |
| VRC (1.2B) vs Loong (7B) on zero-shot text-to-video | VRC · MSR-VTT | CLIPSIM: 0.2903 -> 0.2956 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- VideoPoet — VideoPoet: A Large Language Model for Zero-Shot Video Generation (ICML 2024) · better on 0 of 2 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
- DyCoke: Dynamic Compression of Tokens for Fast Video Large Language Models (CVPR 2025 · similarity 0.92)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.92)
- Less Is More, but Where? Dynamic Token Compression via LLM-Guided Keyframe Prior (NeurIPS 2025 · similarity 0.91)
- Variation-aware Vision Token Dropping for Faster Large Vision-Language Models (CVPR 2026 · similarity 0.91)
- VoCo-LLaMA: Towards Vision Compression with Large Language Models (CVPR 2025 · similarity 0.91)
- PVC: Progressive Visual Token Compression for Unified Image and Video Processing in Large Vision-Language Models (CVPR 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.
