{"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/deepcorrect-correcting-dnn-models-against","title":"DeepCorrect: Correcting DNN models against Image Distortions","arxiv_id":"1705.02406","date":"2017-05-05","proceeding":null,"authors":["Tejas Borkar","Lina Karam"],"abstract":"In recent years, the widespread use of deep neural networks (DNNs) has\nfacilitated great improvements in performance for computer vision tasks like\nimage classification and object recognition. In most realistic computer vision\napplications, an input image undergoes some form of image distortion such as\nblur and additive noise during image acquisition or transmission. Deep networks\ntrained on pristine images perform poorly when tested on such distortions. In\nthis paper, we evaluate the effect of image distortions like Gaussian blur and\nadditive noise on the activations of pre-trained convolutional filters. We\npropose a metric to identify the most noise susceptible convolutional filters\nand rank them in order of the highest gain in classification accuracy upon\ncorrection. In our proposed approach called DeepCorrect, we apply small stacks\nof convolutional layers with residual connections, at the output of these\nranked filters and train them to correct the worst distortion affected filter\nactivations, whilst leaving the rest of the pre-trained filter outputs in the\nnetwork unchanged. Performance results show that applying DeepCorrect models\nfor common vision tasks like image classification (ImageNet), object\nrecognition (Caltech-101, Caltech-256) and scene classification (SUN-397),\nsignificantly improves the robustness of DNNs against distorted images and\noutperforms other alternative approaches..","url_abs":"http://arxiv.org/abs/1705.02406v5","url_pdf":"http://arxiv.org/pdf/1705.02406v5.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":"deepcorrect-correcting-dnn-models-against","repo_url":"https://github.com/tsborkar/DeepCorrect","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"classification-1","task_name":"Classification"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"object-recognition","task_name":"Object Recognition"},{"task_slug":"scene-classification","task_name":"Scene Classification"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}