Research map / Probabilistic and scientific modeling / Variational and generative sampling
The Curse of Conditions: Analyzing and Improving Optimal Transport for Conditional Flow-Based Generation
Ho Kei Cheng, Alexander Schwing
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 conditional weighting to the OT cost (C2OT instead of plain OT) | minibatch OT coupling · 8gaussians->moons binary conditions, adaptive | W2^2: 0.462 -> 0.018 | Improved |
| add conditional weighting to the OT cost (C2OT instead of plain OT) | minibatch OT coupling · 8gaussians->moons continuous conditions, adaptive | W2^2: 2.143 -> 0.013 | Improved |
| use C2OT instead of FM | flow matching · CIFAR-10 class-conditioned, Euler-2 | FID: 105.481 -> 64.694 | Improved |
| use C2OT instead of FM | flow matching · CIFAR-10 class-conditioned, adaptive | FID: 2.922 -> 2.904 | No clear effect |
| use OT instead of FM in conditional generation | flow matching · CIFAR-10 class-conditioned, adaptive | FID: 2.922 -> 4.144 | Hurt |
| use plain OT instead of FM | flow matching · ImageNet-256 latent, adaptive | FID: 3.498 -> 5.726 | Hurt |
| increase OT batch size from 128 to 5120 (ablation) | OT · CIFAR-10, Euler-2 | FID: 64.305 -> 54.966 | Improved |
| increase OT batch size from 128 to 5120 (ablation) | C2OT · CIFAR-10, Euler-2 | FID: 74.517 -> 56.011 | Improved |
| increase OT batch size from 128 to 5120 (ablation) | C2OT · CIFAR-10, adaptive | FID: 2.880 -> 2.945 | Hurt |
| increase OT batch size from 128 to 5120 (ablation) | OT · CIFAR-10, adaptive | FID: 4.142 -> 5.352 | Hurt |
| increase target ratio from 0.01 to 0.1 (ablation) | C2OT · ImageNet-32 caption-conditioned, Euler-2 | FID: 102.380 -> 82.456 | Improved |
| increase target ratio from 0.01 to 0.1 (ablation) | C2OT · ImageNet-256 caption-conditioned, adaptive | FID: 3.290 -> 4.570 | Hurt |
Closest papers by meaning
- Dynamic Conditional Optimal Transport through Simulation-Free Flows (NeurIPS 2024 · similarity 0.90)
- Pairwise Optimal Transports for Training All-to-All Flow-Based Condition Transfer Model (NeurIPS 2025 · similarity 0.89)
- Flow Matching from Viewpoint of Proximal Operators (NeurIPS 2026 · similarity 0.89)
- Optimal Flow Matching: Learning Straight Trajectories in Just One Step (NeurIPS 2024 · similarity 0.87)
- Faster Inference of Flow-Based Generative Models via Improved Data-Noise Coupling (ICLR 2025 · similarity 0.86)
- Computing Optimal Transport Maps and Wasserstein Barycenters Using Conditional Normalizing Flows (ICML 2025 · similarity 0.86)
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.
