Research map / Image generation / Autoregressive and latent generation
Improved Masked Image Generation with Knowledge-Augmented Token Representations
Guotao Liang, Baoquan Zhang, Zhiyuan Wen, Zihao Han, Yunming Ye
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 knowledge-augmented token graph priors (KA) | MaskGIT · ImageNet 256x256 class-conditional | FID: 6.18 -> 5.69 | Improved |
| add knowledge-augmented token graph priors (KA) | MaskGIT · ImageNet 256x256 class-conditional | IS: 182.1 -> 170.2 | Hurt |
| add KA graph priors | AutoNAT · ImageNet 256x256 class-conditional | FID: 2.68 -> 2.45 | Improved |
| add KA graph priors | TiTok-s128 · ImageNet 256x256 class-conditional | FID: 1.97 -> 1.90 | Improved |
| add KA graph priors to shorter-token-sequence model | TiTok-b64 · ImageNet 256x256 class-conditional | FID: 2.48 -> 2.40 | Improved |
| add Gco and Gs graph priors | MaskGen · MS-COCO 30K text-to-image | FID: 22.27 -> 21.01 | Improved |
| add semantic similarity graph alone (ablation) | AutoNAT · ImageNet 256x256 (graph ablation) | FID: 2.68 -> 2.49 | Improved |
| add incompatibility graph alone (ablation) | AutoNAT · ImageNet 256x256 (graph ablation) | FID: 2.68 -> 2.51 | Improved |
| add co-occurrence graph alone (ablation) | AutoNAT · ImageNet 256x256 (graph ablation) | FID: 2.68 -> 2.51 | Improved |
| combine all three graphs vs best single graph (ablation) | AutoNAT · ImageNet 256x256 (graph ablation) | FID: 2.49 -> 2.45 | Improved |
| add all three graphs (ablation) | AutoNAT · ImageNet 256x256 (graph ablation) | IS: 278.8 -> 274.1 | Mixed |
| add incompatibility graph (ablation) | AutoNAT · Token ranking | NDCG@100: 0.0282 -> 0.0284 (mask 0.3), 0.0288 -> 0.0289 (mask 0.5) | No clear effect |
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 2 of 2 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
- Image Understanding Makes for A Good Tokenizer for Image Generation (NeurIPS 2024 · similarity 0.88)
- TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation (CVPR 2025 · similarity 0.86)
- UniTok: a Unified Tokenizer for Visual Generation and Understanding (NeurIPS 2025 · similarity 0.86)
- ENAT: Rethinking Spatial-temporal Interactions in Token-based Image Synthesis (NeurIPS 2024 · similarity 0.86)
- SAM-Guided Masked Token Prediction for 3D Scene Understanding (NeurIPS 2024 · similarity 0.86)
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.85)
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.
