{"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/the-treasure-beneath-convolutional-layers","title":"The Treasure beneath Convolutional Layers: Cross-convolutional-layer Pooling for Image Classification","arxiv_id":"1411.7466","date":"2014-11-27","proceeding":"CVPR 2015 6","authors":["Lingqiao Liu","Chunhua Shen","Anton Van Den Hengel"],"abstract":"A number of recent studies have shown that a Deep Convolutional Neural\nNetwork (DCNN) pretrained on a large dataset can be adopted as a universal\nimage description which leads to astounding performance in many visual\nclassification tasks. Most of these studies, if not all, adopt activations of\nthe fully-connected layer of a DCNN as the image or region representation and\nit is believed that convolutional layer activations are less discriminative.\n  This paper, however, advocates that if used appropriately convolutional layer\nactivations can be turned into a powerful image representation which enjoys\nmany advantages over fully-connected layer activations. This is achieved by\nadopting a new technique proposed in this paper called\ncross-convolutional-layer pooling. More specifically, it extracts subarrays of\nfeature maps of one convolutional layer as local features and pools the\nextracted features with the guidance of feature maps of the successive\nconvolutional layer. Compared with exising methods that apply DCNNs in the\nlocal feature setting, the proposed method is significantly faster since it\nrequires much fewer times of DCNN forward computation. Moreover, it avoids the\ndomain mismatch issue which is usually encountered when applying fully\nconnected layer activations to describe local regions. By applying our method\nto four popular visual classification tasks, it is demonstrated that the\nproposed method can achieve comparable or in some cases significantly better\nperformance than existing fully-connected layer based image representations\nwhile incurring much lower computational cost.","url_abs":"http://arxiv.org/abs/1411.7466v1","url_pdf":"http://arxiv.org/pdf/1411.7466v1.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":"the-treasure-beneath-convolutional-layers","repo_url":"https://github.com/shoaibahmed/CrossLayerPooling","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":null,"task_name":"Image Description"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1411.7466","atlas_url":"https://app.syntology.ai/?focus=1411.7466","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}