Search:
Match:
1 results

Analysis

This paper introduces DeFloMat, a novel object detection framework that significantly improves the speed and efficiency of generative detectors, particularly for time-sensitive applications like medical imaging. It addresses the latency issues of diffusion-based models by leveraging Conditional Flow Matching (CFM) and approximating Rectified Flow, enabling fast inference with a deterministic approach. The results demonstrate superior accuracy and stability compared to existing methods, especially in the few-step regime, making it a valuable contribution to the field.
Reference

DeFloMat achieves state-of-the-art accuracy ($43.32\% ext{ } AP_{10:50}$) in only $3$ inference steps, which represents a $1.4 imes$ performance improvement over DiffusionDet's maximum converged performance ($31.03\% ext{ } AP_{10:50}$ at $4$ steps).