Research map / Image generation / Fast diffusion sampling
JSON: Jigsaw Self-play Optimization for Normalizing Flows
Fengxiang Yang, Tianyi Zheng, Jinwei Chen, Bo 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 jigsaw head to SimFlow (ablation) | SimFlow · ImageNet-1K 256x256 (w/ guidance) | IS: 276.80 -> 284.27 | Improved |
| add jigsaw head to SimFlow (ablation) | SimFlow · ImageNet-1K 256x256 (w/ guidance) | FID: 2.15 -> 2.37 | Hurt |
| replace original self-play with bounded self-play (with jigsaw head) (ablation) | NF with jigsaw head · ImageNet-1K 256x256 (w/ guidance) | training stability: N/A -> 2.19 | Improved |
| add bounded self-play without jigsaw head (ablation) | SimFlow · ImageNet-1K 256x256 (w/ guidance) | IS: 276.80 -> 311.64 | Improved |
| add bounded self-play without jigsaw head (ablation) | SimFlow · ImageNet-1K 256x256 (w/ guidance) | FID: 2.15 -> 2.46 | Hurt |
| add bounded self-play to Jigsaw NF (ablation) | Jigsaw NF · ImageNet-1K 256x256 (w/ guidance) | IS: 284.27 -> 291.85 | Improved |
| add jigsaw-guided self-play (Ljsp) instead of Lbsp (ablation) | Jigsaw NF with bounded self-play · ImageNet-1K 256x256 (w/ guidance) | FID: 2.19 -> 2.10 | Improved |
| add jigsaw-guided self-play (Ljsp) instead of Lbsp (ablation) | Jigsaw NF with bounded self-play · ImageNet-1K 256x256 (w/ guidance) | IS: 291.85 -> 306.20 | Improved |
| use conv projection jigsaw head instead of MLP (ablation) | JSON · ImageNet-1K 256x256 (w/ guidance) | FID: 2.28 -> 2.10 | Improved |
| use conv projection jigsaw head instead of MLP (ablation) | JSON · ImageNet-1K 256x256 (w/ guidance) | IS: 294.03 -> 306.20 | Improved |
| reduce jigsaw patch size from 4 to 1 (ablation) | JSON · ImageNet-1K 256x256 (w/ guidance) | FID: 2.47 -> 2.10 | Improved |
| reduce jigsaw patch size from 4 to 1 (ablation) | JSON · ImageNet-1K 256x256 (w/ guidance) | IS: 290.54 -> 306.20 | 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.
- StarFlow — STARFlow: Scaling Latent Normalizing Flows for High-resolution Image Synthesis (NeurIPS 2025) · 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
- Guided Trajectory Generation with Diffusion Models for Offline Model-based Optimization (NeurIPS 2024 · similarity 0.85)
- Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning (NeurIPS 2025 · similarity 0.84)
- SCoT: Unifying Consistency Models and Rectified Flows via Straight-Consistent Trajectories (NeurIPS 2025 · similarity 0.84)
- Faster Inference of Flow-Based Generative Models via Improved Data-Noise Coupling (ICLR 2025 · similarity 0.84)
- Rectifying Categorical Flows on Statistical Manifolds for One-Step Generation (NeurIPS 2026 · similarity 0.84)
- Optimal Flow Matching: Learning Straight Trajectories in Just One Step (NeurIPS 2024 · similarity 0.84)
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.
