Research map / Image generation / Autoregressive and latent generation
Masked Diffusion Transformer is a Strong Image Synthesizer
Shanghua Gao, Pan Zhou, Ming-Ming Cheng, Shuicheng Yan
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 mask latent modeling with asymmetric structure (MDT) to DiT | DiT-S/2 · ImageNet 256x256 | FID-50K: 68.40 -> 53.46 (400k steps) | Improved |
| use MDT-XL/2 instead of DiT-XL/2 | DiT-XL/2 · ImageNet 256x256 | FID-50K: 9.62 (7000k) vs 9.60 (1300k) | Improved |
| use asymmetric architecture that processes only unmasked tokens in encoder (ablation) | MDT-S/2 · ImageNet 256x256 | FID-50K: 51.56 -> 50.26 | Improved |
| add side-interpolater (ablation) | MDT-S/2 · ImageNet 256x256 | FID-50K: 51.60 -> 50.26 | Improved |
| train on both full and unmasked latents instead of unmasked only (ablation) | MDT-S/2 · ImageNet 256x256 | FID-50K: 76.63 -> 50.26 | Improved |
| compute loss on all tokens instead of masked tokens only (ablation) | MDT-S/2 · ImageNet 256x256 | FID-50K: 58.35 -> 50.26 | Improved |
| add relative positional bias (ablation) | MDT-S/2 · ImageNet 256x256 | FID-50K: 53.56 -> 50.26 | Improved |
| increase side-interpolater blocks from 1 to 3 (ablation) | MDT-S/2 · ImageNet 256x256 | FID-50K: 50.26 -> 51.96 | Hurt |
| raise mask ratio from 0.3 to 0.8 (ablation) | MDT-S/2 · ImageNet 256x256 | FID-50K: 50.26 -> 53.73 | Hurt |
| add Adan optimizer (ablation) | MDT-S/2 · ImageNet 256x256 | FID-50K: 50.26 -> 49.71 | Improved |
| add Min-SNR loss weighting (ablation) | MDT-S/2 with Adan · ImageNet 256x256 | FID-50K: 49.71 -> 45.06 | Improved |
| add encoder long-shortcuts and decoder dense input-shortcuts (ablation) | MDTv2-S/2 · ImageNet 256x256 | FID-50K: 43.82 -> 41.12 -> 38.41 | 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.
- MaskGIT — MaskGIT: Masked Generative Image Transformer (CVPR 2022) · 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
- EDT: An Efficient Diffusion Transformer Framework Inspired by Human-like Sketching (NeurIPS 2024 · similarity 0.91)
- MC-DiT: Contextual Enhancement via Clean-to-Clean Reconstruction for Masked Diffusion Models (NeurIPS 2024 · similarity 0.90)
- Masked Autoencoders Are Effective Tokenizers for Diffusion Models (ICML 2025 · similarity 0.90)
- Cross-view Masked Diffusion Transformers for Person Image Synthesis (ICML 2024 · similarity 0.89)
- SD-DiT: Unleashing the Power of Self-supervised Discrimination in Diffusion Transformer (CVPR 2024 · similarity 0.89)
- LaVin-DiT: Large Vision Diffusion Transformer (CVPR 2025 · similarity 0.89)
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.
