.png){kind=icon}
.png){kind=icon}
Return
A Generic Class-agnostic Object Counting Network with Adaptive Offset Deformable Convolution
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131310.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Class-agnostic object counting (CAC) aims at counting the number of objects in the unseen category in an image. Traditional methods often focus on a single setting, such as Few-shot Counting (FSC) and Zero-shot Counting (ZSC), and they struggle to handle objects with varying scales within the same image and fail to utilize self-similarity cues for accurate localization and counting. In this paper, we design a generic class-agnostic object counting network with Adaptive Offset Deformable Convolution (AODC), which initially focuses on the Reference-less class-agnostic object Counting task without any exemplar, and can be easily extended to FSC and ZSC. Our method calculates the self-similarity maps of the image features and performs 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. We further extend 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. Code is available at https://github.com/WJWu20/AODC .
Keywords:
class-agnostic object counting
adaptive offset deformable convolution
self-similarity
few-shot counting
zero-shot counting
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W