Spatially-Aware Class-Agnostic Object Counting
Computer Vision and Pattern Recognition
Summary
The authors developed UpCount, a method to count objects in images that works better for complex shapes by keeping track of their spatial layout. They improved image features using a vision transformer and special layers to keep details at different scales. UpCount then finds repeating patterns and creates a map showing where objects are to produce an accurate count. They tested it on standard datasets and showed it works well for counting various objects, including vehicles.
Authors
Robert Wijaya, Md. Tanvir Hossain, Amanda Kau, Ngai-Man Cheung
Abstract
Generalised object counting aims to estimate the number of instances of an arbitrary object category from a single image, but many recent methods can struggle on structurally complex objects due to limited spatial modelling. We present \textit{UpCount}, a class-agnostic counter designed to better preserve spatial structure. UpCount strengthens the visual representation by extracting multi-layer features from a ViT-B/16 encoder and reassembling them into a refined multi-scale pyramid that is spatially refined using Dense Prediction Transformers and FeatUp, yielding features with improved structural and spatial sensitivity; a proposal--verification counting head then identifies repeated patterns and produces a density map for the final count. On FSC-147, UpCount achieves 12.39 MAE and 100.89 RMSE on the test set, and it transfers effectively to vehicle counting on CARPK (6.27 MAE, 8.79 RMSE). Code: https://github.com/r28112072-rgb/upcount