Research map / Video / Video generation
FastSTAR: Spatiotemporal Token Pruning for Efficient Autoregressive Video Synthesis
Sungwoong Yune, Suheon Jeong, Joo-Young Kim
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 spatiotemporal token pruning with partial update (20/30/40/70%) | InfinityStar · T2V 720p | end-to-end latency / speedup: 81.7 s baseline -> 2.01x speedup | Improved |
| spatiotemporal pruning instead of image-based reduction (SparseVAR, FastVAR, ToMe) | InfinityStar · T2V 720p | PSNR: FastSTAR 28.29 (highest) vs baselines lower; baselines speedup 1.44x-1.76x | Improved |
| apply FastSTAR to image-to-video | InfinityStar · I2V 720p | PSNR: 25.65 (best among reduction methods) | Improved |
| add temporal similarity to spatial-only pruning (ablation) | FastSTAR · T2V | quality metrics | Improved |
| use random pruning instead of similarity-based pruning (ablation) | FastSTAR · T2V | quality metrics | Hurt |
| norm degree p=2 vs p=1 or infinity in score fusion (ablation) | FastSTAR · T2V | SSIM / LPIPS | Improved |
| prune final-scale tokens beyond 70% (ablation) | FastSTAR · T2V scale 29 | quality | Hurt |
| token pruning instead of token merging (40% ratio) (ablation) | FastSTAR · STAR feature maps | MSE | Improved |
| use FastSTAR at 480p vs 720p | FastSTAR · T2V/I2V/V2V 480p | speedup | Mixed |
Closest papers by meaning
- InfinityStar: Unified Spacetime AutoRegressive Modeling for Visual Generation (NeurIPS 2025 · similarity 0.90)
- VideoMAR: Autoregressive Video Generation with Continuous Tokens (NeurIPS 2025 · similarity 0.89)
- StarGen: A Spatiotemporal Autoregression Framework with Video Diffusion Model for Scalable and Controllable Scene Generation (CVPR 2025 · similarity 0.88)
- Parallelized Autoregressive Visual Generation (CVPR 2025 · similarity 0.88)
- Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis (CVPR 2024 · similarity 0.88)
- Autoregressive Video Generation without Vector Quantization (ICLR 2025 · similarity 0.87)
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.
