Research map / Video / Video generation
Error Analyses of Auto-Regressive Video Diffusion Models
Jing Wang, Fengzhuo Zhang, Xiaoli Li, Vincent Y.~ F. Tan, Tianyu Pang, Chao Du, Aixin Sun, Zhuoran Yang
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 |
|---|---|---|---|
| increase history length from 16 to 112 frames (ablation) | Prepending AR-VDM · DMLab history retrieval | SRR: 0.18 -> 0.33 | Improved |
| increase history length from 16 to 112 frames (ablation) | Prepending AR-VDM · Minecraft history retrieval | SSIM: 0.50 -> 0.62 | Improved |
| increase compression budget from 8 to 48 history latents (ablation) | Prepending AR-VDM · DMLab history retrieval | SRR: 0.17 -> 0.25 | Improved |
| increase compression budget from 8 to 48 history latents (ablation) | Prepending AR-VDM · Minecraft history retrieval | SSIM: 0.51 -> 0.56 | Improved |
| increase history length from 16 to 112 frames (ablation) | Channel Concat AR-VDM · DMLab history retrieval | SRR: 0.09 -> 0.11 | Improved |
| use prepending instead of channel concatenation | AR-VDM history conditioning · DMLab history retrieval (history 48) | SRR: 0.12 -> 0.25 | Improved |
| add soft memory compression instead of last 16 frames | Prepending AR-VDM · DMLab history retrieval, overall | SRR: 0.260 -> 0.305 | Improved |
| add soft memory compression instead of FIFO | Prepending AR-VDM · Minecraft history retrieval, overall | SSIM: 0.405 -> 0.552 | Improved |
| add soft memory compression instead of last 32 frames with skip | Prepending AR-VDM · Minecraft history retrieval, overall | SSIM: 0.472 -> 0.552 | Improved |
| increase history length from 16 to 112 (ablation) | Prepending AR-VDM · DMLab FVD | FVD: 302.59 -> 308.81 | No clear effect |
| increase compression budget from 8 to 48 (ablation) | Prepending AR-VDM · DMLab frames 48-64 | PSNR: 7.88 -> 6.76 | Hurt |
| increase compression budget from 8 to 48 (ablation) | Prepending AR-VDM · Minecraft frames 48-64 | PSNR: 10.29 -> 9.47 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- StreamingT2V — StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025) · better on 2 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
- AR-Diffusion: Asynchronous Video Generation with Auto-Regressive Diffusion (CVPR 2025 · similarity 0.91)
- Ca2-VDM: Efficient Autoregressive Video Diffusion Model with Causal Generation and Cache Sharing (ICML 2025 · similarity 0.89)
- History-Guided Video Diffusion (ICML 2025 · similarity 0.89)
- Identifying and Solving Conditional Image Leakage in Image-to-Video Diffusion Model (NeurIPS 2024 · similarity 0.89)
- Video-Mirai: Autoregressive Video Diffusion Models Need Foresight (NeurIPS 2026 · similarity 0.89)
- TRIP: Temporal Residual Learning with Image Noise Prior for Image-to-Video Diffusion Models (CVPR 2024 · similarity 0.88)
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.
