Research map / Image generation / Fast diffusion sampling
Align Your Trajectory Tangent: Training Better Consistency Models via Manifold-Aligned Tangents
Beomsu Kim, ByungHee Cha, Jong Chul Ye
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 AYT manifold feature distance loss | ECT · CIFAR10 (unconditional), 1-step | FID: 3.60 -> 2.61 | Improved |
| add AYT manifold feature distance loss | ECT · CIFAR10 (unconditional), 2-step | FID: 2.11 vs. 2.13 | No clear effect |
| add AYT manifold feature distance loss | ECT-S · ImageNet 64x64 (class-conditional), 1-step | FID: 5.51 -> 4.41 | Improved |
| add AYT manifold feature distance loss | ECT-S · ImageNet 64x64 (class-conditional), 2-step | FID: 3.27 vs. 3.18 | No clear effect |
| train manifold features with more transformation groups (degradation, geometric, color) (ablation) | AYT · CIFAR10 (unconditional) | FID | Improved |
| add geometric transformations to manifold feature training (ablation) | AYT · CIFAR10 (unconditional) | FID | Improved |
| use manifold feature distance instead of LPIPS (ablation) | CM training · CIFAR10 (unconditional) | FID (1 and 2 step) | Improved |
| use manifold feature distance instead of LPIPS (ablation) | CM training · CIFAR10 (unconditional) | denoising FID at t=0.8 | Improved |
| reduce batch size to 16 with AYT (ablation) | AYT · CIFAR10 (unconditional) | FID | No clear effect |
Closest papers by meaning
- Consistency Trajectory Models: Learning Probability Flow ODE Trajectory of Diffusion (ICLR 2024 · similarity 0.89)
- SCoT: Unifying Consistency Models and Rectified Flows via Straight-Consistent Trajectories (NeurIPS 2025 · similarity 0.88)
- Riemannian Consistency Model (NeurIPS 2025 · similarity 0.87)
- Metric Flow Matching for Smooth Interpolations on the Data Manifold (NeurIPS 2024 · similarity 0.87)
- Categorical Flow Matching on Statistical Manifolds (NeurIPS 2024 · similarity 0.87)
- Truncated Consistency Models (ICLR 2025 · similarity 0.87)
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.
