Dense center-direction regression for object counting and localization with point supervision

Open Webpage

Domen Tabernik, Jon Muhovic, Danijel Skocaj

Published: 01 Jan 2024, Last Modified: 05 Mar 2025Pattern Recognit. 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•CeDiRNet learns center-direction regression based on point supervision.•Center-directions point towards object center providing a large support for detection.•Mapping images into center-direction space enables domain-agnostic localization.•Localization network can be learned from synthetic data.•CeDiRNet achieves SOTA results on different problem domains.