AI Paper Atlas

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

ECCV 2026 · Image generation › Fast diffusion sampling · approach: Guidance and alignment for flow matching · 5 citations (OpenAlex)

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 changedWhereMeasuredEffect
add RL (Q-learning) stage scheduler (ablation)CACFM · Aesthetic, 4-stepLAION Aesthetic: 5.430 -> 5.893Improved
add DMD loss (ablation)CACFM · Aesthetic, 4-stepLAION Aesthetic: 5.485 -> 5.893Improved
use RL scheduler instead of loss-proportional EMA sampler (ablation)CFM · FLUX 4-stepHPS: 0.236 -> 0.276Improved
use Q-learning instead of epsilon-greedy bandit (ablation)CACFM · FLUX 4-stepHPS: 0.242 -> 0.276Improved
use Logit-Normal instead of uniform sampling (ablation)Naive CFM · 4-stepAvg Rank: 4.11 -> 4.89Hurt
set number of stages M=4 rather than 3 (ablation)CACFM · FLUX 4-stepFID: 39.85 -> 39.19Improved
set number of stages M=4 rather than 6 (ablation)CACFM · FLUX 4-stepFID: 39.51 -> 39.19Improved
use CACFM instead of PCM under equal wall-clockPCM · CC3M FID, 24hFID: 43.10 -> 40.82Improved
increase inference steps from 4 to 8CACFM · FLUX CC3MFID: 39.19 -> 36.96Improved
increase inference steps from 4 to 16PCM · SDXL CC3MFID: 37.26 -> 40.47Hurt
add RL agent and DMD per-iteration overheadCACFM · training timerelative time per step: 1.00x -> 1.18xHurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy