Research map / Image generation / Autoregressive and latent generation
PixelU: A U-Shaped Transformer for Efficient End-to-End Pixel Diffusion
Zipeng Guo, Lichen Ma, Yu He, Xiaolong Fu, Jingling Fu, Junshi Huang, Yan Li
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 skip connections to JiT-B/16 (x-pred) (ablation) | JiT-B/16 · ImageNet 256 160ep cfg=1 | FID: 43.65 -> 38.53 | Improved |
| add DDT decoder (x-pred) (ablation) | JiT-B/16 · ImageNet 256 160ep cfg=1 | FID: 43.65 -> 43.48 | No clear effect |
| add PixelDiT decoder instead of skips (ablation) | JiT-B/16 · ImageNet 256 160ep cfg=1 | FID: 43.65 -> 39.20 | Improved |
| switch from x-pred to v-pred with skips (ablation) | JiT-B/16 + skips · ImageNet 256 160ep cfg=1 | FID: 38.53 -> 185.54 | Hurt |
| add DeCo decoder under v-pred (ablation) | JiT-B/16 · ImageNet 256 160ep cfg=1 | FID: 190.11 -> 42.07 | Improved |
| add spatial down-sampling to skip connections (ablation) | PixelU-B/16 · ImageNet 256 160ep cfg=1 | FID: 38.53 -> 34.92 | Improved |
| add down-sampling without skip connections (ablation) | JiT-B/16 · ImageNet 256 160ep cfg=1 | FID: 43.65 -> 48.87 | Hurt |
| balanced [4,4,4] block allocation vs [3,6,3] (ablation) | PixelU-B/16 · ImageNet 256 160ep cfg=1 | FID: 38.08 -> 34.92 | Improved |
| use one down-sampling stage instead of two (ablation) | PixelU-B/16 · ImageNet 256 160ep cfg=1 | FID: 42.70 -> 34.92 | Improved |
| add in-context class tokens | PixelU-B/16 · ImageNet 256 160ep cfg=1 | FID: 34.92 -> 30.88 | Improved |
| add REPA to class tokens | PixelU-B/16 · ImageNet 256 160ep cfg=1 | FID: 30.88 -> 26.86 | Improved |
| add CFG and CFG interval, 600 epochs | PixelU-B/16 · ImageNet 256 | FID: 26.86 -> 3.40 | Improved |
Closest papers by meaning
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.91)
- PixelDiT2: Representation-Grounded Pixel Diffusion Transformers (NeurIPS 2026 · similarity 0.90)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.89)
- Diffusion Models Without Attention (CVPR 2024 · similarity 0.88)
- Rethinking Cross-Layer Information Routing in Diffusion Transformer (NeurIPS 2026 · similarity 0.88)
- Dual-Rate Diffusion: Accelerating diffusion models with an interleaved heavy-light network (NeurIPS 2026 · similarity 0.88)
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.
