Contrastive Point Feature Matching for Open-world Object Counting


Ngo Xuan Cuong (University of Arkansas)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)Supplementary

Abstract

Open-world object counting refers to estimating the number of target objects specified by a text description. Current state-of-the-art density-based methods rely on density map regression trained with counting-supervised losses, which often overlook the spatial and visual correspondence between image regions and object instances. Furthermore, the absence of spatial supervision leads to inconsistent performance in challenging scenarios such as crowding and occlusion. We propose CPMNet (Contrastive Point Matching Network), a novel framework that introduces spatial supervision through contrastive learning with point-level guidance. CPMNet regress object point locations from the predicted density map through Sinkhorn EM (expectation maximization), extracts their corresponding features, and applies contrastive learning to enhance discrimination between target objects and irrelevant regions. Experimental results on a benchmark dataset demonstrate that our method offers enhanced performance in open-world counting, providing a more precise and reliable solution for estimating object counts.

Citation

@inproceedings{Cuong_2025_BMVC,
author    = {Ngo Xuan Cuong},
title     = {Contrastive Point Feature Matching for Open-world Object Counting},
booktitle = {36th British Machine Vision Conference 2025, {BMVC} 2025, Sheffield, UK, November 24-27, 2025},
publisher = {BMVA},
year      = {2025},
url       = {https://bmva-archive.org.uk/bmvc/2025/assets/papers/Paper_1183/paper.pdf}
}


Copyright © 2025 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection