Research map / Video / Video generation
MAGVIT: Masked Generative Video Transformer
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 MAGVIT-L-CG rather than TATS | UCF-101 class-conditional generation · UCF-101 | FVD: 332±18 -> 76±2 | Helped |
| use MAGVIT-L-CG rather than TATS | UCF-101 class-conditional generation · UCF-101 | Inception Score: 79.28±0.38 -> 89.27±0.15 | Helped |
| use MAGVIT-L-FP rather than RaMViD | frame prediction · BAIR | FVD: 84 -> 62 | Helped |
| use MAGVIT-L-FP rather than Video Diffusion | frame prediction · Kinetics-600 | FVD: 16.2±0.3 -> 9.9±0.3 | Helped |
| add multi-task training rather than single-task frame prediction training | MAGVIT-B · BAIR frame prediction | FVD: 201.1 -> 32.8 | Helped |
| add Lrefine to Lmask + Lrecons (ablation) | COMMIT · BAIR frame prediction | FVD: 51 -> 48 | Helped |
| add Lrefine to Lmask + Lrecons (ablation) | COMMIT · BAIR eight-task generation | FVD: 53 -> 33 | Helped |
| add Lrecons and Lrefine to Lmask (ablation) | COMMIT · BAIR frame prediction | FVD: 388 -> 48 | Helped |
| add Lrecons and Lrefine to Lmask (ablation) | COMMIT · BAIR eight-task generation | FVD: 143 -> 33 | Helped |
| use COMMIT decoding rather than MaskGIT decoding with 3D-VQ (ablation) | MAGVIT · BAIR frame prediction | FVD: 122 (74) -> 76 (48) | Helped |
| use non-autoregressive COMMIT decoding rather than autoregressive decoding (ablation) | MAGVIT with 3D-VQ · BAIR frame prediction | FVD: 91 (56) -> 76 (48) | Helped |
| use central rather than average inflation (ablation) | MAGVIT 3D-VQ-B with ImageNet initialization · UCF-101 training-set reconstruction | FVD: 103 -> 58 | Helped |
Closest papers by meaning
- Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis (CVPR 2024 · similarity 0.88)
- Masked Generative Nested Transformers with Decode Time Scaling (ICML 2025 · similarity 0.88)
- Mask^2DiT: Dual Mask-based Diffusion Transformer for Multi-Scene Long Video Generation (CVPR 2025 · similarity 0.88)
- Training-Free Efficient Video Generation via Dynamic Token Carving (NeurIPS 2025 · similarity 0.88)
- Video-LaVIT: Unified Video-Language Pre-training with Decoupled Visual-Motional Tokenization (ICML 2024 · similarity 0.87)
- OmniTokenizer: A Joint Image-Video Tokenizer for Visual Generation (NeurIPS 2024 · 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.