Research map / Image generation / Autoregressive and latent generation
MaskGIT: Masked Generative Image Transformer
Check 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 bidirectional masked-token generation with iterative parallel decoding instead of autoregressive raster-order decoding | ImageNet class-conditional synthesis · ImageNet 256×256 | FID: 15.78 -> 6.18 | Helped |
| Use bidirectional masked-token generation with iterative parallel decoding instead of autoregressive raster-order decoding | ImageNet class-conditional synthesis · ImageNet 512×512 | FID: 26.52 -> 7.32 | Helped |
| Use parallel iterative decoding rather than autoregressive decoding | ImageNet generation · ImageNet 256×256 and 512×512 | # steps: 256 -> 8; 1024 -> 12 | Helped |
| Use MaskGIT rather than BigGAN-deep | ImageNet generation · ImageNet 256×256 | Recall: 0.28 -> 0.51 | Helped |
| Use MaskGIT rather than BigGAN-deep | ImageNet generation · ImageNet 512×512 | FID: 8.43 -> 7.32 | Helped |
| Apply classifier-based rejection sampling at 0.05 acceptance rate | MaskGIT · ImageNet 256×256 | IS: 182.1 -> 355.6 | Helped |
| Apply classifier-based rejection sampling at 0.05 acceptance rate | MaskGIT · ImageNet 512×512 | IS: 156.0 -> 342.0 | Helped |
| Use MaskGIT rather than CoModGAN | Places2 outpainting · Places2, right 50% outpainting | FID: 7.80 -> 6.78 | Helped |
| Use MaskGIT rather than CoModGAN | Places2 inpainting · Places2, center 50%×50% inpainting | IS: 21.82 -> 22.95 | Helped |
| Use cosine rather than exponential mask scheduling (ablation) | MaskGIT · ImageNet 256×256 | FID: 7.89 -> 6.06 | Helped |
| Use cosine rather than cubic mask scheduling (ablation) | MaskGIT · ImageNet 256×256 | FID: 7.26 -> 6.06 | Helped |
| Use cosine rather than square mask scheduling (ablation) | MaskGIT · ImageNet 256×256 | FID: 6.35 -> 6.06 | Helped |
Closest papers by meaning
- Halton Scheduler for Masked Generative Image Transformer (ICLR 2025 · similarity 0.92)
- Masked Generative Nested Transformers with Decode Time Scaling (ICML 2025 · similarity 0.87)
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.87)
- Meissonic: Revitalizing Masked Generative Transformers for Efficient High-Resolution Text-to-Image Synthesis (ICLR 2025 · similarity 0.86)
- ENAT: Rethinking Spatial-temporal Interactions in Token-based Image Synthesis (NeurIPS 2024 · similarity 0.86)
- MaskINT: Video Editing via Interpolative Non-autoregressive Masked Transformers (CVPR 2024 · 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.