{"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/spatial-frequency-loss-for-learning","title":"Spatial Frequency Loss for Learning Convolutional Autoencoders","arxiv_id":"1806.02336","date":"2018-06-06","proceeding":null,"authors":["Naoyuki Ichimura"],"abstract":"This paper presents a learning method for convolutional autoencoders (CAEs)\nfor extracting features from images. CAEs can be obtained by utilizing\nconvolutional neural networks to learn an approximation to the identity\nfunction in an unsupervised manner. The loss function based on the pixel loss\n(PL) that is the mean squared error between the pixel values of original and\nreconstructed images is the common choice for learning. However, using the loss\nfunction leads to blurred reconstructed images. A method for learning CAEs\nusing a loss function computed from features reflecting spatial frequencies is\nproposed to mitigate the problem. The blurs in reconstructed images show lack\nof high spatial frequency components mainly constituting edges and detailed\ntextures that are important features for tasks such as object detection and\nspatial matching. In order to evaluate the lack of components, a convolutional\nlayer with a Laplacian filter bank as weights is added to CAEs and the mean\nsquared error of features in a subband, called the spatial frequency loss\n(SFL), is computed from the outputs of each filter. The learning is performed\nusing a loss function based on the SFL. Empirical evaluation demonstrates that\nusing the SFL reduces the blurs in reconstructed images.","url_abs":"http://arxiv.org/abs/1806.02336v1","url_pdf":"http://arxiv.org/pdf/1806.02336v1.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":"spatial-frequency-loss-for-learning","repo_url":"https://github.com/nram812/SuperResolutionSatellite","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"object-detection","task_name":"Object Detection"},{"task_slug":"object-detection-1","task_name":"object-detection"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1806.02336","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}