Research map / Image generation / Autoregressive and latent generation
DiP: Taming Diffusion Models in Pixel Space
Zhennan Chen, Junwei Zhu, Xu Chen, Jiangning Zhang, Xiaobin Hu, Hanzhen Zhao, Chengjie Wang, Jian Yang, Ying Tai
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 a conv U-Net Patch Detailer Head to the DiT backbone (ablation) | DiP (DiT, patch 16) · ImageNet 256x256 | FID: 5.28 -> 2.16 | Improved |
| deepen DiT-only from 26 to 32 layers instead of adding head (ablation) | DiT-only · ImageNet 256x256 | FID: 5.28 -> 4.91 | Improved |
| widen DiT-only hidden dim to 1536 (ablation) | DiT-only · ImageNet 256x256 | FID: 5.28 -> 2.83 | Improved |
| use standard MLP as the patch head (ablation) | Patch Detailer Head · ImageNet 256x256 | FID: 2.16 -> 6.92 | Hurt |
| use intra-patch attention instead of conv U-Net head (ablation) | Patch Detailer Head · ImageNet 256x256 | FID: 2.16 -> 2.98 | Hurt |
| use coordinate-based MLP instead of conv U-Net head (ablation) | Patch Detailer Head · ImageNet 256x256 | FID: 2.16 -> 2.20 | Hurt |
| place head intermediately or hybrid rather than post-hoc (ablation) | Patch Detailer Head placement · ImageNet 256x256 | FID | Hurt |
| train in pixel space with DiP rather than latent DiT-XL | DiP-XL/16 · ImageNet 512x512 | FID: 3.04 -> 2.31 | Improved |
| increase head depth (down/up stages) (ablation) | Patch Detailer Head · ImageNet 256x256 | FID | Improved |
| increase head width (ablation) | Patch Detailer Head · ImageNet 256x256 | FID | No clear effect |
Closest papers by meaning
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.92)
- PixelDiT2: Representation-Grounded Pixel Diffusion Transformers (NeurIPS 2026 · similarity 0.91)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.90)
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 2025 · similarity 0.90)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.90)
- Diffusion Models Without Attention (CVPR 2024 · 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.
