Research map / Image generation / Editing and style transfer
The Image Local Autoregressive Transformer
Chenjie Cao, Yuxin Hong, Xiang Li, Chengrong Wang, Chengming Xu, Yanwei Fu, Xiangyang Xue
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 |
|---|---|---|---|
| Use iLAT with local discrete representations and the LA attention mask instead of Taming | locally guided pose synthesis · Penn Action (PA) | PSNR: 21.43 -> 22.94 | Improved |
| Use iLAT with local discrete representations and the LA attention mask instead of Taming | locally guided pose synthesis · Penn Action (PA) | FID: 33.53 -> 27.36 | Improved |
| Use iLAT instead of Taming | locally guided image synthesis · Synthetic DeepFashion (SDF) | SSIM: 0.539 -> 0.599 | Improved |
| Use iLAT instead of Taming | locally guided image synthesis · Synthetic DeepFashion (SDF) | FID: 72.77 -> 70.58 | Improved |
| Generate only masked regions autoregressively | iLAT · PA | inference time (sec/image): 8.551 -> 3.426 | Improved |
| Generate only masked regions autoregressively | iLAT · FFHQ | inference time (sec/image): 8.183 -> 1.180 | Improved |
| Add two-stream convolutions to prevent encoder information leakage (ablation) | iLAT · PA | FID: 31.83 -> 27.36 | Improved |
| Add two-stream convolutions to prevent encoder information leakage (ablation) | iLAT · PA | PSNR: 21.68 -> 22.94 | Improved |
| Add two-stream convolutions to prevent encoder information leakage (ablation) | iLAT · PA | SSIM: 0.748 -> 0.800 | Improved |
| Add two-stream convolutions to prevent encoder information leakage (ablation) | iLAT · PA | MAE: 0.056 -> 0.046 | Improved |
| Use the LA attention mask with Taming rather than its vanilla attention mask | Taming · Face editing | average human evaluation score: 0.33 -> 3.33 | 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.
- SC-FEGAN — SC-FEGAN: Face Editing Generative Adversarial Network With User's Sketch and Color (ICCV 2019) · 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
- Fundamental Limits of Visual Autoregressive Transformers: Universal Approximation Abilities (ICML 2025 · similarity 0.87)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.86)
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.86)
- A Spark of Vision-Language Intelligence: 2-Dimensional Autoregressive Transformer for Efficient Finegrained Image Generation (ICLR 2025 · similarity 0.86)
- Revisiting Non-Autoregressive Transformers for Efficient Image Synthesis (CVPR 2024 · similarity 0.85)
- EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 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.
