{"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/learning-non-lambertian-object-intrinsics","title":"Learning Non-Lambertian Object Intrinsics across ShapeNet Categories","arxiv_id":"1612.08510","date":"2016-12-27","proceeding":"CVPR 2017 7","authors":["Jian Shi","Yue Dong","Hao Su","Stella X. Yu"],"abstract":"We consider the non-Lambertian object intrinsic problem of recovering diffuse\nalbedo, shading, and specular highlights from a single image of an object.\n  We build a large-scale object intrinsics database based on existing 3D models\nin the ShapeNet database. Rendered with realistic environment maps, millions of\nsynthetic images of objects and their corresponding albedo, shading, and\nspecular ground-truth images are used to train an encoder-decoder CNN. Once\ntrained, the network can decompose an image into the product of albedo and\nshading components, along with an additive specular component.\n  Our CNN delivers accurate and sharp results in this classical inverse problem\nof computer vision, sharp details attributed to skip layer connections at\ncorresponding resolutions from the encoder to the decoder. Benchmarked on our\nShapeNet and MIT intrinsics datasets, our model consistently outperforms the\nstate-of-the-art by a large margin.\n  We train and test our CNN on different object categories. Perhaps surprising\nespecially from the CNN classification perspective, our intrinsics CNN\ngeneralizes very well across categories. Our analysis shows that feature\nlearning at the encoder stage is more crucial for developing a universal\nrepresentation across categories.\n  We apply our synthetic data trained model to images and videos downloaded\nfrom the internet, and observe robust and realistic intrinsics results. Quality\nnon-Lambertian intrinsics could open up many interesting applications such as\nimage-based albedo and specular editing.","url_abs":"http://arxiv.org/abs/1612.08510v1","url_pdf":"http://arxiv.org/pdf/1612.08510v1.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":"learning-non-lambertian-object-intrinsics","repo_url":"https://github.com/shi-jian/shapenet-intrinsics","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"torch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"decoder","task_name":"Decoder"},{"task_slug":"object","task_name":"Object"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1612.08510","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}