{"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-based-learning-using-reflection-and","title":"CNN based Learning using Reflection and Retinex Models for Intrinsic Image Decomposition","arxiv_id":"1712.01056","date":"2017-12-04","proceeding":"CVPR 2018 6","authors":["Anil S. Baslamisli","Hoang-An Le","Theo Gevers"],"abstract":"Most of the traditional work on intrinsic image decomposition rely on\nderiving priors about scene characteristics. On the other hand, recent research\nuse deep learning models as in-and-out black box and do not consider the\nwell-established, traditional image formation process as the basis of their\nintrinsic learning process. As a consequence, although current deep learning\napproaches show superior performance when considering quantitative benchmark\nresults, traditional approaches are still dominant in achieving high\nqualitative results. In this paper, the aim is to exploit the best of the two\nworlds. A method is proposed that (1) is empowered by deep learning\ncapabilities, (2) considers a physics-based reflection model to steer the\nlearning process, and (3) exploits the traditional approach to obtain intrinsic\nimages by exploiting reflectance and shading gradient information. The proposed\nmodel is fast to compute and allows for the integration of all intrinsic\ncomponents. To train the new model, an object centered large-scale datasets\nwith intrinsic ground-truth images are created. The evaluation results\ndemonstrate that the new model outperforms existing methods. Visual inspection\nshows that the image formation loss function augments color reproduction and\nthe use of gradient information produces sharper edges. Datasets, models and\nhigher resolution images are available at https://ivi.fnwi.uva.nl/cv/retinet.","url_abs":"http://arxiv.org/abs/1712.01056v2","url_pdf":"http://arxiv.org/pdf/1712.01056v2.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":[],"tasks":[{"task_slug":"deep-learning","task_name":"Deep Learning"},{"task_slug":"intrinsic-image-decomposition","task_name":"Intrinsic Image Decomposition"}],"methods":[],"datasets_introduced":[{"slug":"shapenet-intrinsic-images-v1-0","name":"ShapeNet Intrinsic Images v1.0","full_name":""}],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1712.01056","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}