DDMCB: Open-world object detection empowered by Denoising Diffusion Models and Calibration Balance

Open Webpage

Yangyang Huang, Xing Xi, Ronghua Luo

Published: 01 Jan 2025, Last Modified: 17 Apr 2025Image Vis. Comput. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•We propose an unsupervised method using a denoising diffusion model.•It tackles label bias and information bottlenecks in reconstruction.•We introduce a probability balancing method for label bias.•Results beat prior top methods on OWOD detection benchmark.