Research map / Video / Video-language understanding
InfoTok: Adaptive Discrete Video Tokenizer via Information-Theoretic Compression
Haotian Ye, Qiyuan He, Jiaqi Han, Puheng Li, Jiaojiao Fan, Zekun Hao, Fitsum Reda, Yogesh Balaji, Huayu Chen, Sheng Liu, Angela Yao, James Zou and 3 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 |
|---|---|---|---|
| use InfoTok instead of ElasticTok at BPP16 0.81 | tokenizer · TokenBench-256x256 | FVD: 141 -> 49 | Improved |
| use InfoTok instead of ElasticTok at BPP16 0.56 | tokenizer · TokenBench-256x256 | PSNR: 27.34 -> 29.27 | Improved |
| use InfoTok instead of ElasticTok at BPP16 0.56 | tokenizer · DAVIS-256x256 | FVD: 930 -> 540 | Improved |
| use InfoTok-Flex instead of InfoTok at BPP16 0.56 | tokenizer · DAVIS-256x256 | PSNR: 24.52 -> 24.84 | No clear effect |
| use ELBO-based router instead of exhaustive optimal search (ablation) | InfoTok-Flex · TokenBench-256x256 BPP16 0.56 | PSNR: 29.39 -> 29.30 | No clear effect |
| use ELBO-based router instead of exhaustive optimal search (ablation) | InfoTok-Flex · TokenBench-256x256 BPP16 0.31 | FVD: 165 -> 155 | No clear effect |
| use ELBO-based compressor instead of R2L masking (ablation) | InfoTok · TokenBench BPP16 0.56 | PSNR: 27.43 -> 29.30 | Improved |
| use ELBO-based compressor instead of Jump masking (ablation) | InfoTok · TokenBench BPP16 0.56 | FVD: 84 -> 71 | Improved |
| use ELBO-based compressor instead of Jump masking (ablation) | InfoTok · TokenBench BPP16 0.56 | LPIPS: 0.173 -> 0.179 | Hurt |
| use ELBO adaptive mechanism instead of uniform on ViT backbone (ablation) | ElasticTok ViT backbone · TokenBench BPP16 0.56 | FVD: 198 -> 114 | Improved |
| use InfoTok mechanism instead of ElasticTok mechanism on Cosmos | Cosmos · inference latency per video | latency: 13.45 s -> 1.23 s | Improved |
| use Cosmos + InfoTok at BPP16 0.56 instead of Cosmos at 1.00 | Cosmos Arch · TokenBench 360p | FVD: 27 -> 56 | Hurt |
Closest papers by meaning
- Single-pass Adaptive Image Tokenization for Minimum Program Search (NeurIPS 2025 · similarity 0.89)
- VQToken: Neural Discrete Token Representation Learning for Extreme Token Reduction in Video Large Language Models (NeurIPS 2025 · similarity 0.88)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.88)
- ElasticTok: Adaptive Tokenization for Image and Video (ICLR 2025 · similarity 0.87)
- PVC: Progressive Visual Token Compression for Unified Image and Video Processing in Large Vision-Language Models (CVPR 2025 · similarity 0.87)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · 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.
