Research map / Image generation / Fast diffusion sampling
Dual-Solver: A Generalized ODE Solver for Diffusion Models with Dual Prediction
Soochul Park, Yeon Ju Lee
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 second-order corrector to first-order predictor (p1 -> p1c2) (ablation) | Dual-Solver · DiT, NFE=3 | cross-entropy loss: 0.667 -> 0.574 | Improved |
| use second-order predictor with corrector (p2c2) instead of p1c2 (ablation) | Dual-Solver · DiT, NFE=3 | cross-entropy loss: 0.574 -> 5.009 | Hurt |
| make all of gamma, tau, kappa learnable instead of fixing gamma=-1 (ablation) | Dual-Solver · DiT, NFE=5 | cross-entropy loss: 7.676 -> 0.197 | Improved |
| make all parameters learnable instead of kappa=0 (remove second-order residual) (ablation) | Dual-Solver · DiT, NFE=3 | cross-entropy loss: 0.944 -> 0.574 | Improved |
| make all parameters learnable instead of fixing or sharing them (ablation) | Dual-Solver · DiT, NFE=9 | cross-entropy loss: 0.173 vs 0.169 (tau,kappa shared) | Mixed |
| use log1p domain change (Type 2) instead of linear/log interpolation (Type 1) (ablation) | Dual-Solver · DiT, NFE=9 | FID: 3.64 -> 2.67 | Improved |
| use log1p domain change (Type 2) instead of Type 1 (ablation) | Dual-Solver · DiT, NFE=3 | FID: 23.38 -> 24.91 | Hurt |
| learn with hard-label classification instead of sample regression (ablation) | Dual-Solver · DiT, NFE=3 | FID: 107.13 -> 24.91 | Improved |
| learn with hard-label classification instead of feature regression (VGG) (ablation) | Dual-Solver · DiT, NFE=5 | FID: 5.48 -> 3.52 | Improved |
| learn with hard labels instead of soft labels (ablation) | Dual-Solver · DiT, NFE=5 | FID: 4.90 -> 3.52 | Improved |
| use trajectory-regression objective (Dual-Solver-R) instead of classification (Dual-Solver-C) (ablation) | Dual-Solver · GM-DiT, NFE=9 | FID: 2.60 -> 2.32 (R better at NFE 8-9) | Hurt |
| replace DS-Solver with Dual-Solver | DiT · ImageNet, NFE=3 | FID: 67.31 -> 24.91 | Improved |
Closest papers by meaning
- PFDiff: Training-Free Acceleration of Diffusion Models Combining Past and Future Scores (ICLR 2025 · similarity 0.91)
- TADA: Improved Diffusion Sampling with Training-free Augmented DynAmics (NeurIPS 2025 · similarity 0.91)
- Dual-Rate Diffusion: Accelerating diffusion models with an interleaved heavy-light network (NeurIPS 2026 · similarity 0.90)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.90)
- Distilling ODE Solvers of Diffusion Models into Smaller Steps (CVPR 2024 · similarity 0.90)
- DEFT: Efficient Fine-tuning of Diffusion Models by Learning the Generalised $h$-transform (NeurIPS 2024 · similarity 0.90)
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.
