Research map / Image generation / Fast diffusion sampling
Curvature-Adaptive Consistency Flow Matching: Autonomous Trajectory Optimization via Reinforcement Learning
Songtao Tian, Guhan Chen, Bohan Li, Jingyi Ma, Zixiong Yu
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 RL (Q-learning) stage scheduler (ablation) | CACFM · Aesthetic, 4-step | LAION Aesthetic: 5.430 -> 5.893 | Improved |
| add DMD loss (ablation) | CACFM · Aesthetic, 4-step | LAION Aesthetic: 5.485 -> 5.893 | Improved |
| use RL scheduler instead of loss-proportional EMA sampler (ablation) | CFM · FLUX 4-step | HPS: 0.236 -> 0.276 | Improved |
| use Q-learning instead of epsilon-greedy bandit (ablation) | CACFM · FLUX 4-step | HPS: 0.242 -> 0.276 | Improved |
| use Logit-Normal instead of uniform sampling (ablation) | Naive CFM · 4-step | Avg Rank: 4.11 -> 4.89 | Hurt |
| set number of stages M=4 rather than 3 (ablation) | CACFM · FLUX 4-step | FID: 39.85 -> 39.19 | Improved |
| set number of stages M=4 rather than 6 (ablation) | CACFM · FLUX 4-step | FID: 39.51 -> 39.19 | Improved |
| use CACFM instead of PCM under equal wall-clock | PCM · CC3M FID, 24h | FID: 43.10 -> 40.82 | Improved |
| increase inference steps from 4 to 8 | CACFM · FLUX CC3M | FID: 39.19 -> 36.96 | Improved |
| increase inference steps from 4 to 16 | PCM · SDXL CC3M | FID: 37.26 -> 40.47 | Hurt |
| add RL agent and DMD per-iteration overhead | CACFM · training time | relative time per step: 1.00x -> 1.18x | Hurt |
Closest papers by meaning
- TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment (NeurIPS 2026 · similarity 0.88)
- Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data (NeurIPS 2025 · similarity 0.88)
- PRECISE: SDE-Consistent Stochastic Sampling for RL Post-Training of Flow-Matching Models (NeurIPS 2026 · similarity 0.88)
- Flow Matching Policy Optimization with Mirror Descent and Entropy Constraints (NeurIPS 2026 · similarity 0.87)
- Elucidating the Preconditioning in Consistency Distillation (ICLR 2025 · similarity 0.87)
- FlowPolicy: Enabling Fast and Robust 3D Flow-Based Policy via Consistency Flow Matching for Robot Manipulation (AAAI 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.
