{"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/deepfirearm-learning-discriminative-feature","title":"DeepFirearm: Learning Discriminative Feature Representation for Fine-grained Firearm Retrieval","arxiv_id":"1806.02984","date":"2018-06-08","proceeding":null,"authors":["Jiedong Hao","Jing Dong","Wei Wang","Tieniu Tan"],"abstract":"There are great demands for automatically regulating inappropriate appearance\nof shocking firearm images in social media or identifying firearm types in\nforensics. Image retrieval techniques have great potential to solve these\nproblems. To facilitate research in this area, we introduce Firearm 14k, a\nlarge dataset consisting of over 14,000 images in 167 categories. It can be\nused for both fine-grained recognition and retrieval of firearm images. Recent\nadvances in image retrieval are mainly driven by fine-tuning state-of-the-art\nconvolutional neural networks for retrieval task. The conventional single\nmargin contrastive loss, known for its simplicity and good performance, has\nbeen widely used. We find that it performs poorly on the Firearm 14k dataset\ndue to: (1) Loss contributed by positive and negative image pairs is unbalanced\nduring training process. (2) A huge domain gap exists between this dataset and\nImageNet. We propose to deal with the unbalanced loss by employing a double\nmargin contrastive loss. We tackle the domain gap issue with a two-stage\ntraining strategy, where we first fine-tune the network for classification, and\nthen fine-tune it for retrieval. Experimental results show that our approach\noutperforms the conventional single margin approach by a large margin (up to\n88.5% relative improvement) and even surpasses the strong triplet-loss-based\napproach.","url_abs":"http://arxiv.org/abs/1806.02984v2","url_pdf":"http://arxiv.org/pdf/1806.02984v2.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":"deepfirearm-learning-discriminative-feature","repo_url":"https://github.com/jdhao/deep_firearm","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"image-retrieval","task_name":"Image Retrieval"},{"task_slug":"retrieval","task_name":"Retrieval"},{"task_slug":null,"task_name":"Triplet"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}