Research map / Video / Video generation
Improving Autoregressive Video Modeling with History Understanding
Wenyang Luo, Haina Qin, Bing Li, Jiwen Lu, Xin Tao, Pengfei Wan, Kun Gai
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 masked history modeling (MiMo) to the ACDiT baseline (ablation) | ACDiT-B · Kinetics-600 (DiT-B, 100K steps) | FVD: 54.814 -> 36.601 | Improved |
| add MiMo on top of REPA-Both (ablation) | REPA-Both · Kinetics-600 (DiT-B) | FVD: 36.542 -> 34.133 | Improved |
| add REPA feature distillation on history frames only | ACDiT-B · Kinetics-600 (DiT-B) | FVD: 54.814 -> 40.022 | Improved |
| add REPA on both history and future frames instead of future only | ACDiT-B + REPA · Kinetics-600 (DiT-B) | FVD: 40.253 -> 36.542 | Improved |
| use history+future REPA instead of only future frames (accuracy) | ACDiT-B + REPA · Kinetics-600 linear probing | accuracy: 17.04 -> 19.23 | Improved |
| predict next frame in addition to current frame as masked-history target (ablation) | MiMo · Kinetics-600 (DiT-B) | FVD: 41.832 -> 36.601 | Improved |
| predict current frame in addition to next frame as masked-history target (ablation) | MiMo · Kinetics-600 (DiT-B) | FVD: 37.782 -> 36.601 | Improved |
| add NextNext frame target (ablation) | MiMo · Kinetics-600 (DiT-B) | FVD: 36.601 -> 36.263 | No clear effect |
| place decoder after layer 11 instead of 12 (ablation) | MiMo · Kinetics-600 (DiT-B) | FVD: 36.601 -> 35.815 | No clear effect |
| place decoder after layer 9 instead of 12 (ablation) | MiMo · Kinetics-600 (DiT-B) | FVD: 36.601 -> 37.593 | Hurt |
| add RoPE and separate LayerNorm for history frames (ablation) | Vanilla DiT · Kinetics-600 (DiT-B) | FVD: 37.763 -> 36.601 | Improved |
| set lambda to 0.5 instead of 0.1 (ablation) | MiMo · Kinetics-600 (DiT-B) | FVD: 40.213 -> 36.601 | Improved |
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.
- Phenaki — Phenaki: Variable Length Video Generation from Open Domain Textual Descriptions (ICLR 2023) · better on 1 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
- History-Guided Video Diffusion (ICML 2025 · similarity 0.90)
- Video-Mirai: Autoregressive Video Diffusion Models Need Foresight (NeurIPS 2026 · similarity 0.88)
- Ca2-VDM: Efficient Autoregressive Video Diffusion Model with Causal Generation and Cache Sharing (ICML 2025 · similarity 0.87)
- Addressable Memory for Video World Models (NeurIPS 2026 · similarity 0.87)
- CanvasMAR: Improving Masked Autoregressive Video Prediction With Canvas (NeurIPS 2026 · similarity 0.86)
- VideoMAR: Autoregressive Video Generation with Continuous Tokens (NeurIPS 2025 · 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.
