A Generic Class-agnostic Object Counting Network with Adaptive Offset Deformable Convolution

Yuanwu Xu, Haofeng Zhang

Keywords: Class-agnostic Object Counting, Adaptive Offset, 4D Convolution, Deformable Convolution

Abstract: Class-agnostic object counting (CAC) aims at counting the number of objects in the unseen category in an image. In this paper, we design a generic class-agnostic object counting network with Adaptive Offset Deformable Convolution (AODC), which initially focus on the reference-less class-agnostic object counting task without any exemplar. Our method calculates the self-similarity maps of the image features and performing a 4D convolution on these maps, obtaining the adaptive offsets for the deformable convolution, so that the model can obtain complete information about the object at that location. Through this process, AODC is able to recognize objects of different scales in a same sample. In addition to this, we adopt our approach to both zero-shot setting and few-shot setting, the former with semantic text and the latter with visual exemplars as references. We conduct experiments on the few-shot object counting dataset FSC-147, as well as other large-scale datasets, and show that our method significantly outperforms state-of-the-art approaches on all the three settings.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 12962