{"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/cnn-fixations-an-unraveling-approach-to","title":"CNN Fixations: An unraveling approach to visualize the discriminative image regions","arxiv_id":"1708.06670","date":"2017-08-22","proceeding":null,"authors":["Konda Reddy Mopuri","Utsav Garg","R. Venkatesh Babu"],"abstract":"Deep convolutional neural networks (CNN) have revolutionized various fields\nof vision research and have seen unprecedented adoption for multiple tasks such\nas classification, detection, captioning, etc. However, they offer little\ntransparency into their inner workings and are often treated as black boxes\nthat deliver excellent performance. In this work, we aim at alleviating this\nopaqueness of CNNs by providing visual explanations for the network's\npredictions. Our approach can analyze variety of CNN based models trained for\nvision applications such as object recognition and caption generation. Unlike\nexisting methods, we achieve this via unraveling the forward pass operation.\nProposed method exploits feature dependencies across the layer hierarchy and\nuncovers the discriminative image locations that guide the network's\npredictions. We name these locations CNN-Fixations, loosely analogous to human\neye fixations.\n  Our approach is a generic method that requires no architectural changes,\nadditional training or gradient computation and computes the important image\nlocations (CNN Fixations). We demonstrate through a variety of applications\nthat our approach is able to localize the discriminative image locations across\ndifferent network architectures, diverse vision tasks and data modalities.","url_abs":"http://arxiv.org/abs/1708.06670v3","url_pdf":"http://arxiv.org/pdf/1708.06670v3.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":"cnn-fixations-an-unraveling-approach-to","repo_url":"https://github.com/utsavgarg/cnn-fixations","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"cnn-fixations-an-unraveling-approach-to","repo_url":"https://github.com/val-iisc/cnn-fixations","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"caption-generation","task_name":"Caption Generation"},{"task_slug":"image-captioning","task_name":"Image Captioning"},{"task_slug":"object-recognition","task_name":"Object Recognition"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}