{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/class-agnostic-counting","title":"Class-Agnostic Counting","arxiv_id":"1811.00472","date":"2018-11-01","proceeding":null,"authors":["Erika Lu","Weidi Xie","Andrew Zisserman"],"abstract":"Nearly all existing counting methods are designed for a specific object\nclass. Our work, however, aims to create a counting model able to count any\nclass of object. To achieve this goal, we formulate counting as a matching\nproblem, enabling us to exploit the image self-similarity property that\nnaturally exists in object counting problems. We make the following three\ncontributions: first, a Generic Matching Network (GMN) architecture that can\npotentially count any object in a class-agnostic manner; second, by\nreformulating the counting problem as one of matching objects, we can take\nadvantage of the abundance of video data labeled for tracking, which contains\nnatural repetitions suitable for training a counting model. Such data enables\nus to train the GMN. Third, to customize the GMN to different user\nrequirements, an adapter module is used to specialize the model with minimal\neffort, i.e. using a few labeled examples, and adapting only a small fraction\nof the trained parameters. This is a form of few-shot learning, which is\npractical for domains where labels are limited due to requiring expert\nknowledge (e.g. microbiology). We demonstrate the flexibility of our method on\na diverse set of existing counting benchmarks: specifically cells, cars, and\nhuman crowds. The model achieves competitive performance on cell and crowd\ncounting datasets, and surpasses the state-of-the-art on the car dataset using\nonly three training images. When training on the entire dataset, the proposed\nmethod outperforms all previous methods by a large margin.","url_abs":"http://arxiv.org/abs/1811.00472v1","url_pdf":"http://arxiv.org/pdf/1811.00472v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"class-agnostic-counting","repo_url":"https://github.com/erikalu/class-agnostic-counting","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"crowd-counting","task_name":"Crowd Counting"},{"task_slug":"few-shot-learning","task_name":"Few-Shot Learning"},{"task_slug":"object","task_name":"Object"},{"task_slug":"object-counting","task_name":"Object Counting"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1811.00472","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}