Research map / Image generation / Fast diffusion sampling
GAS: Improving Discretization of Diffusion ODEs via Generalized Adversarial Solver
Aleksandr Oganov, Ilya Bykov, Eva Neudachina, Mishan Aliev, Alexander Tolmachev, Alexander Sidorov, Zuev Aleksandr, Andrey Okhotin, Denis Rakitin, Aibek Alanov
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 adversarial loss to distillation (GS -> GAS) (ablation) | Generalized Solver · FFHQ 64x64 NFE=4 | FID: 10.70 -> 7.86 | Improved |
| add adversarial loss to distillation (GS -> GAS) (ablation) | Generalized Solver · ImageNet-256 NFE=4 | FID: 7.87 -> 5.38 | Improved |
| add adversarial loss to distillation (GS -> GAS) (ablation) | Generalized Solver · LSUN-Bedroom NFE=4 | FID: 9.83 -> 6.68 | Improved |
| add adversarial loss to distillation (GS -> GAS) (ablation) | Generalized Solver · AFHQv2 NFE=10 | FID: 2.25 -> 2.31 | Hurt |
| add adversarial loss to distillation (GS -> GAS) (ablation) | Generalized Solver · CIFAR10 NFE=10 | FID: 2.18 -> 2.17 | No clear effect |
| add adversarial loss to distillation (GS -> GAS) (ablation) | Generalized Solver · MS-COCO SD v1.5 NFE=6 | FID: 11.71 -> 11.73 | No clear effect |
| use Generalized Solver parameterization instead of S4S Alt | solver distillation · CIFAR10 NFE=4 | FID: 6.35 -> 4.41 | Improved |
| use Generalized Solver parameterization instead of S4S Alt | solver distillation · ImageNet-256 NFE=4 | FID: 5.13 -> 7.87 | Hurt |
| add theoretical solver guidance to coefficients (ablation) | Generalized Solver · FFHQ | FID | Improved |
| train on 1400 instead of 49000 samples (ablation) | Generalized Solver · CIFAR10 NFE=4 | FID: 4.39 -> 4.35 | No clear effect |
| train GAS on 1400 instead of 5000 samples (ablation) | GAS · FFHQ NFE=4 | FID: 7.83 -> 9.44 | Hurt |
| train solver on AFHQv2 and apply to FFHQ (ablation) | GAS · FFHQ NFE=4 | FID: 7.86 -> 9.39 | Mixed |
Closest papers by meaning
- Distilling ODE Solvers of Diffusion Models into Smaller Steps (CVPR 2024 · similarity 0.90)
- AdjointDEIS: Efficient Gradients for Diffusion Models (NeurIPS 2024 · similarity 0.90)
- Revisiting Diffusion Models: From Generative Pre-training to One-Step Generation (ICML 2025 · similarity 0.89)
- S4S: Solving for a Fast Diffusion Model Solver (ICML 2025 · similarity 0.89)
- Learning to Solve Generative ODEs Beyond the Linear Span (NeurIPS 2026 · similarity 0.89)
- Bespoke Solvers for Generative Flow Models (ICLR 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.
