Research map / Image generation / Autoregressive and latent generation
PixelFolder: An Efficient Progressive Pixel Synthesis Network for Image Generation
Jing He, Yiyi Zhou, Qi Zhang, Jun Peng, Yunhang Shen, Xiaoshuai Sun, Chao Chen, Rongrong Ji
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 pixel folding/unfolding instead of down-sampling+DeConv (ablation) | PixelFolder · FFHQ 256x256 (200k-step ablation) | FID: 8.36 -> 5.49 | Improved |
| use unfolding instead of DeConv after folding (ablation) | PixelFolder · FFHQ 256x256 (200k-step ablation) | FID: 5.60 -> 5.49 | Improved |
| use folding instead of down-sampling (both with DeConv) (ablation) | PixelFolder · FFHQ 256x256 (200k-step ablation) | FID: 5.53 -> 5.60 | No clear effect |
| use pixel unfolding instead of DeConv, reducing params and GMACs (ablation) | PixelFolder · FFHQ 256x256 (200k-step ablation) | GMACs: 86.53 -> 23.78 | Improved |
| use pixel unfolding instead of DeConv (ablation) | PixelFolder · FFHQ 256x256 (200k-step ablation) | Recall: 0.371 -> 0.514 | Improved |
| add coordinate embeddings in first stage (ablation) | PixelFolder · LSUN Church 256x256 (200k-step ablation) | FID: 4.95 -> 4.78 | Improved |
| add multi-stage connection (ablation) | PixelFolder · LSUN Church 256x256 (200k-step ablation) | FID: 5.46 -> 4.78 | Improved |
| add multi-stage connection (ablation) | PixelFolder · LSUN Church 256x256 (200k-step ablation) | Precision: 0.532 -> 0.602 | Improved |
| use PixelFolder instead of CIPS | PixelFolder · LSUN Church 256x256 | FID: 2.92 -> 2.45 | Improved |
| use PixelFolder instead of CIPS | PixelFolder · Efficiency (256x256) | GMACs: 223.36 -> 23.78 | Improved |
| use PixelFolder instead of StyleGAN2 | PixelFolder · FFHQ 256x256 | FID: 3.83 -> 3.77 | Improved |
| use PixelFolder instead of StyleGAN2 | PixelFolder · FFHQ 1024x1024 | FID: 2.84 -> 2.98 | Hurt |
Closest papers by meaning
- PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining (NeurIPS 2026 · similarity 0.86)
- PixelDiT2: Representation-Grounded Pixel Diffusion Transformers (NeurIPS 2026 · similarity 0.86)
- L2P: Unlocking Latent Potential for Pixel Generation (NeurIPS 2026 · similarity 0.85)
- Is One GPU Enough? Pushing Image Generation at Higher-Resolutions with Foundation Models. (NeurIPS 2024 · similarity 0.85)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.85)
- Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass Diffusion Transformers (ICML 2024 · 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.
