{"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/mildnet-a-lightweight-single-scaled-deep","title":"MILDNet: A Lightweight Single Scaled Deep Ranking Architecture","arxiv_id":"1903.00905","date":"2019-03-03","proceeding":null,"authors":["Anirudha Vishvakarma"],"abstract":"Multi-scale deep CNN architecture [1, 2, 3] successfully captures both fine\nand coarse level image descriptors for visual similarity task, but they come up\nwith expensive memory overhead and latency. In this paper, we propose a\ncompeting novel CNN architecture, called MILDNet, which merits by being vastly\ncompact (about 3 times). Inspired by the fact that successive CNN layers\nrepresent the image with increasing levels of abstraction, we compressed our\ndeep ranking model to a single CNN by coupling activations from multiple\nintermediate layers along with the last layer. Trained on the famous\nStreet2shop dataset [4], we demonstrate that our approach performs as good as\nthe current state-of-the-art models with only one third of the parameters,\nmodel size, training time and significant reduction in inference time. The\nsignificance of intermediate layers on image retrieval task has also been shown\nto be performing on popular datasets Holidays, Oxford, Paris [5]. So even\nthough our experiments are done on ecommerce domain, it is applicable to other\ndomains as well. We further did an ablation study to validate our hypothesis by\nchecking the impact on adding each intermediate layer. With this we also\npresent two more useful variants of MILDNet, a mobile model (12 times smaller)\nfor on-edge devices and a compactly featured model (512-d feature embeddings)\nfor systems with less RAMs and to reduce the ranking cost. Further we present\nan intuitive way to automatically create a tailored in-house triplet training\ndataset, which is very hard to create manually. This solution too can also be\ndeployed as an all-inclusive visual similarity solution. Finally, we present\nour entire production level architecture which currently powers visual\nsimilarity at Fynd.","url_abs":"http://arxiv.org/abs/1903.00905v2","url_pdf":"http://arxiv.org/pdf/1903.00905v2.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":"mildnet-a-lightweight-single-scaled-deep","repo_url":"https://github.com/gofynd/mildnet","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"mildnet-a-lightweight-single-scaled-deep","repo_url":"https://github.com/Ducvoccer/mildnet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"fine-grained-visual-recognition","task_name":"Fine-Grained Visual Recognition"},{"task_slug":"image-retrieval","task_name":"Image Retrieval"},{"task_slug":"product-recommendation","task_name":"Product Recommendation"},{"task_slug":"recommendation-systems","task_name":"Recommendation Systems"},{"task_slug":"retrieval","task_name":"Retrieval"},{"task_slug":null,"task_name":"Triplet"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-retrieval-on-street2shop-topwear","task":"Image Retrieval","dataset":"street2shop - topwear","model":"MILDNet","rank_in_archive_order":2,"of":2,"metrics":{"Accuracy":"93.69"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}