{"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-visual-classification-via-2","title":"Fine-Grained Visual Classification via Internal Ensemble Learning Transformer","arxiv_id":null,"date":"2023-02-13","proceeding":"IEEE Transactions on Multimedia 2023 2","authors":["Qin Xu","Jiahui Wang","Bo Jiang","Bin Luo"],"abstract":"Recently, vision transformers (ViTs) have been investigated in fine-grained visual recognition (FGVC) and are now\r\nconsidered state of the art. However, most ViT-based works ignore\r\nthe different learning performances of the heads in the multihead self-attention (MHSA) mechanism and its layers. To address\r\nthese issues, in this paper, we propose a novel internal ensemble\r\nlearning transformer (IELT) for FGVC. The proposed IELT\r\ninvolves three main modules: multi-head voting (MHV) module,\r\ncross-layer refinement (CLR) module, and dynamic selection (DS)\r\nmodule. To solve the problem of the inconsistent performances of\r\nmultiple heads, we propose the MHV module, which considers\r\nall of the heads in each layer as weak learners and votes for\r\ntokens of discriminative regions as cross-layer feature based on\r\nthe attention maps and spatial relationships. To effectively mine\r\nthe cross-layer feature and suppress the noise, the CLR module\r\nis proposed, where the refined feature is extracted and the assist\r\nlogits operation is developed for the final prediction. In addition,\r\na newly designed DS module adjusts the token selection number\r\nat each layer by weighting their contributions of the refined\r\nfeature. In this way, the idea of ensemble learning is combined\r\nwith the ViT to improve fine-grained feature representation. The\r\nexperiments demonstrate that our method achieves competitive\r\nresults compared with the state of the art on five popular FGVC\r\ndatasets. Source code has been released and can be found at\r\nhttps://github.com/mobulan/IELT.","url_abs":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10042971","url_pdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10042971","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-visual-classification-via-2","repo_url":"https://github.com/mobulan/ielt","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"classification-1","task_name":"Classification"},{"task_slug":"ensemble-learning","task_name":"Ensemble Learning"},{"task_slug":"fine-grained-image-classification","task_name":"Fine-Grained Image Classification"},{"task_slug":"fine-grained-visual-recognition","task_name":"Fine-Grained Visual Recognition"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/fine-grained-image-classification-on-nabirds","task":"Fine-Grained Image Classification","dataset":"NABirds","model":"IELT","rank_in_archive_order":17,"of":30,"metrics":{"Accuracy":"90.8%"},"uses_additional_data":false},{"leaderboard":"/sota/fine-grained-image-classification-on-oxford","task":"Fine-Grained Image Classification","dataset":"Oxford 102 Flowers","model":"IELT","rank_in_archive_order":1,"of":25,"metrics":{"Accuracy":"99.64%"},"uses_additional_data":false},{"leaderboard":"/sota/fine-grained-image-classification-on-oxford-1","task":"Fine-Grained Image Classification","dataset":"Oxford-IIIT Pet Dataset","model":"IELT","rank_in_archive_order":7,"of":15,"metrics":{"Accuracy":"95.28%"},"uses_additional_data":false},{"leaderboard":"/sota/fine-grained-image-classification-on-stanford-1","task":"Fine-Grained Image Classification","dataset":"Stanford Dogs","model":"IELT","rank_in_archive_order":12,"of":24,"metrics":{"Accuracy":"91.8%"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}