{"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/beyond-saliency-understanding-convolutional","title":"Beyond saliency: understanding convolutional neural networks from saliency prediction on layer-wise relevance propagation","arxiv_id":"1712.08268","date":"2017-12-22","proceeding":null,"authors":["Heyi Li","Yunke Tian","Klaus Mueller","Xin Chen"],"abstract":"Despite the tremendous achievements of deep convolutional neural networks\n(CNNs) in many computer vision tasks, understanding how they actually work\nremains a significant challenge. In this paper, we propose a novel two-step\nunderstanding method, namely Salient Relevance (SR) map, which aims to shed\nlight on how deep CNNs recognize images and learn features from areas, referred\nto as attention areas, therein. Our proposed method starts out with a\nlayer-wise relevance propagation (LRP) step which estimates a pixel-wise\nrelevance map over the input image. Following, we construct a context-aware\nsaliency map, SR map, from the LRP-generated map which predicts areas close to\nthe foci of attention instead of isolated pixels that LRP reveals. In human\nvisual system, information of regions is more important than of pixels in\nrecognition. Consequently, our proposed approach closely simulates human\nrecognition. Experimental results using the ILSVRC2012 validation dataset in\nconjunction with two well-established deep CNN models, AlexNet and VGG-16,\nclearly demonstrate that our proposed approach concisely identifies not only\nkey pixels but also attention areas that contribute to the underlying neural\nnetwork's comprehension of the given images. As such, our proposed SR map\nconstitutes a convenient visual interface which unveils the visual attention of\nthe network and reveals which type of objects the model has learned to\nrecognize after training. The source code is available at\nhttps://github.com/Hey1Li/Salient-Relevance-Propagation.","url_abs":"http://arxiv.org/abs/1712.08268v5","url_pdf":"http://arxiv.org/pdf/1712.08268v5.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":"beyond-saliency-understanding-convolutional","repo_url":"https://github.com/Hey1Li/Salient-Relevance-Propagation","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"beyond-saliency-understanding-convolutional","repo_url":"https://github.com/Hey1Li/Saliency-Relevance-Propagation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"saliency-prediction","task_name":"Saliency Prediction"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"grouped-convolution","method_name":"Grouped Convolution"},{"method_slug":"local-response-normalization","method_name":"Local Response Normalization"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1712.08268","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}