Research map / Video / Video segmentation and tracking
MaskViT: Masked Visual Pre-Training for Video Prediction
Agrim Gupta, Stephen Tian, Yunzhi Zhang, Jiajun Wu, Roberto Martín-Martín, Li Fei-Fei
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 |
|---|---|---|---|
| Condition predictions on the goal frame | MaskViT · BAIR | FVD: 93.7 -> 76.9 | Improved |
| Condition predictions on actions | MaskViT · BAIR | FVD: 93.7 -> 70.5 | Improved |
| Use variable training mask ratios rather than fixed ratios (ablation) | MaskViT · BAIR | FVD: 96.6 vs. 110.9 | Improved |
| Increase embedding dimension from 768 to 1024 (ablation) | MaskViT · BAIR | FVD: 96.6 -> 94.2 | Improved |
| Increase transformer blocks from 6 to 8 (ablation) | MaskViT · BAIR | FVD: 96.6 -> 99.3 | Hurt |
| Use the 16×4×4 spatiotemporal window instead of full self-attention (ablation) | MaskViT · BAIR | FVD: 98.2 -> 96.6 | Improved |
| Use fixed mask ratio 0.98 (ablation) | MaskViT · BAIR | FVD: 96.6 -> 214.4 | Hurt |
| Use concave rather than linear or convex mask scheduling (ablation) | MaskViT iterative decoding · Video prediction datasets | FVD | Improved |
| Add temperature-annealed Gumbel noise to token confidence (ablation) | MaskViT iterative decoding · Video prediction datasets | motion diversity | Improved |
| Increase decoding iterations beyond the point where FVD stops improving (ablation) | MaskViT iterative decoding · Video prediction datasets | FVD | No clear effect |
| Add domain-specific finetuning data | MaskViT robot planner · Real robot table setting and sweeping | successes: 1/30 -> 16/30 | Improved |
| Combine domain-specific data during training versus post-finetuning (ablation) | MaskViT robot planner · Real robot table setting and sweeping | successes: 18/30 vs. 16/30 | No clear effect |
Later papers that compared with this method
4 later accepted papers report MaskViT as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- CanvasMAR: Improving Masked Autoregressive Video Prediction With Canvas (NeurIPS 2026 · better on 1 of 1 shared result)
- Pre-Trained Video Generative Models as World Simulators (AAAI 2026 · better on 1 of 1 shared result)
- iVideoGPT: Interactive VideoGPTs are Scalable World Models (NeurIPS 2024 · better on 3 of 3 shared results)
- MOSO: Decomposing MOtion, Scene and Object for Video Prediction (CVPR 2023 · better on 2 of 3 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
- Mask2IV: Interaction-Centric Video Generation via Mask Trajectories (AAAI 2026 · similarity 0.88)
- Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representations (ICML 2025 · similarity 0.87)
- Visual Representation Learning with Stochastic Frame Prediction (ICML 2024 · similarity 0.86)
- Understanding and Enhancing Mask-Based Pretraining towards Universal Representations (NeurIPS 2025 · similarity 0.86)
- Video Language Model Pretraining with Spatio-temporal Masking (CVPR 2025 · similarity 0.86)
- Grounding Video Models to Actions through Goal Conditioned Exploration (ICLR 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.
