Research map / Image generation / Autoregressive and latent generation
PixNerd: Pixel Neural Field Diffusion
Shuai Wang, Ziteng Gao, Chenhui Zhu, Weilin Huang, Limin Wang
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 Adams-2 solver instead of Euler (50 steps) (ablation) | PixNerd-XL/16 · ImageNet 256x256 class-conditional | FID: 2.29 -> 2.16 | Improved |
| increase sampling steps from 50 to 100 (Euler) (ablation) | PixNerd-XL/16 · ImageNet 256x256 class-conditional | FID: 2.29 -> 2.15 | Improved |
| increase sampling steps from 50 to 100 (Euler) (ablation) | PixNerd-XL/16 · ImageNet 512x512 class-conditional | FID: 3.41 -> 2.84 | Improved |
| use Adams-4 solver instead of Euler/Adams-2 (ablation) | PixNerd-XL/16 · ImageNet 256x256 | FID | Hurt |
| increase neural field MLP depth from 1 to 4 layers (ablation) | PixNerd-L/16 · ImageNet 256x256 | FID | Improved |
| increase neural field channels from 36 to 64 (ablation) | PixNerd-L/16 · ImageNet 256x256 | FID | Improved |
| increase neural field channels from 64 to 72 (ablation) | PixNerd-L/16 · ImageNet 256x256 | FID | No clear effect |
| use DCT-basis coordinate encoding instead of sine/cosine (ablation) | PixNerd-L/16 · ImageNet 256x256 | FID | Improved |
| also normalize neural field output features (vs FC1 or FC1/FC2 only) (ablation) | PixNerd-L/16 · ImageNet 256x256 | FID | Improved |
| use pixel neural field decoder instead of linear projection | Baseline-L/16 · ImageNet 256x256 | training loss | Improved |
Closest papers by meaning
- PiD: Fast and High-Resolution Latent Decoding with Pixel Diffusion (NeurIPS 2026 · similarity 0.91)
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.90)
- PixelDiT2: Representation-Grounded Pixel Diffusion Transformers (NeurIPS 2026 · similarity 0.90)
- PixArt-$\alpha$: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis (ICLR 2024 · similarity 0.89)
- PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining (NeurIPS 2026 · similarity 0.88)
- Image Neural Field Diffusion Models (CVPR 2024 · 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.
