{"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/fine-grained-representation-learning-and","title":"Fine-Grained Representation Learning and Recognition by Exploiting Hierarchical Semantic Embedding","arxiv_id":"1808.04505","date":"2018-08-14","proceeding":null,"authors":["Tianshui Chen","Wenxi Wu","Yuefang Gao","Le Dong","Xiaonan Luo","Liang Lin"],"abstract":"Object categories inherently form a hierarchy with different levels of\nconcept abstraction, especially for fine-grained categories. For example, birds\n(Aves) can be categorized according to a four-level hierarchy of order, family,\ngenus, and species. This hierarchy encodes rich correlations among various\ncategories across different levels, which can effectively regularize the\nsemantic space and thus make prediction less ambiguous. However, previous\nstudies of fine-grained image recognition primarily focus on categories of one\ncertain level and usually overlook this correlation information. In this work,\nwe investigate simultaneously predicting categories of different levels in the\nhierarchy and integrating this structured correlation information into the deep\nneural network by developing a novel Hierarchical Semantic Embedding (HSE)\nframework. Specifically, the HSE framework sequentially predicts the category\nscore vector of each level in the hierarchy, from highest to lowest. At each\nlevel, it incorporates the predicted score vector of the higher level as prior\nknowledge to learn finer-grained feature representation. During training, the\npredicted score vector of the higher level is also employed to regularize label\nprediction by using it as soft targets of corresponding sub-categories. To\nevaluate the proposed framework, we organize the 200 bird species of the\nCaltech-UCSD birds dataset with the four-level category hierarchy and construct\na large-scale butterfly dataset that also covers four level categories.\nExtensive experiments on these two and the newly-released VegFru datasets\ndemonstrate the superiority of our HSE framework over the baseline methods and\nexisting competitors.","url_abs":"http://arxiv.org/abs/1808.04505v1","url_pdf":"http://arxiv.org/pdf/1808.04505v1.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":"fine-grained-representation-learning-and","repo_url":"https://github.com/HCPLab-SYSU/HSE","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"fine-grained-image-classification","task_name":"Fine-Grained Image Classification"},{"task_slug":"fine-grained-image-recognition","task_name":"Fine-Grained Image Recognition"},{"task_slug":"representation-learning","task_name":"Representation Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1808.04505","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}