{ "id": 37273, "title": "alternative methods not using deep learning CNN", "url": "/competitions/carvana-image-masking-challenge/discussion/37273", "author_name": "", "post_date": "2017-07-30T06:56:05.359288900Z", "votes": 12, "comment_count": 1, "views": 0, "content": "

Another way to approach the problem is multi-view segmentation or 3d reconstruction. Anyone considering methods like these:

\n\n\n\n

of course it is possible to develop from these ideas and make a deep CNN for it (e.g. using CNN to replace some prediction or model function)

", "messages": [ { "id": "208556", "postDate": "07/30/2017 06:56:05", "content": "

Another way to approach the problem is multi-view segmentation or 3d reconstruction. Anyone considering methods like these:

\n\n\n\n

of course it is possible to develop from these ideas and make a deep CNN for it (e.g. using CNN to replace some prediction or model function)

", "rawMarkdown": "Another way to approach the problem is multi-view segmentation or 3d reconstruction. Anyone considering methods like these:\n\n - https://hal.inria.fr/inria-00568915/document\n\n - http://igl.ethz.ch/projects/light-field-segmentation/\n\n - http://www.oliverwang.info/\n (Efficient 3D Object Segmentation from Densely Sampled Light Fields with Applications to 3D Reconstruction)\n\n - https://www.youtube.com/watch?v=UPS8ZPHPY5A\n\n - http://chenlab.ece.cornell.edu/projects/iModel/\n\n https://www.youtube.com/watch?v=iPsutuvbWbc\n\n - http://www.cv-foundation.org/openaccess/content_iccv_2013/papers/Djelouah_Multi-view_Object_Segmentation_2013_ICCV_paper.pdf\n\n - http://adjelouah.name/index.php/projects/multi-view-segmentation/\n\n - https://www.youtube.com/watch?v=UhwayReG9i8\n\n - https://www.youtube.com/watch?v=VoEBJfI8kDw\n\n - https://www.youtube.com/watch?v=462hKb86CIY\n\n - https://hal.inria.fr/hal-00735718/PDF/Final_N-tuple_Multi-View_Silhouette_Extraction-1.pdf\n\n - https://www.youtube.com/watch?v=LEi5qllwMKg\n\n- https://www.youtube.com/watch?v=skFFsJZ3xCM\n\n- https://www.youtube.com/watch?v=tjMpgno6k5A\n\n- https://www.youtube.com/watch?v=PrUVo3potl4\n\n- http://www.apsipa.org/proceedings_2014/Data/paper/1267.pdf\n\n\nof course it is possible to develop from these ideas and make a deep CNN for it (e.g. using CNN to replace some prediction or model function)", "votes": null }, { "id": "208576", "postDate": "07/30/2017 08:57:43", "content": "

Yeah, I thought about reconstructing 3D model of a car and background. As far as I know those methods have very high precision (in submillimeter range sometimes).

", "rawMarkdown": "Yeah, I thought about reconstructing 3D model of a car and background. As far as I know those methods have very high precision (in submillimeter range sometimes).", "votes": null } ], "comments": [ { "id": 208576, "author_name": "ceperaang", "author_url": "", "post_date": "07/30/2017 08:57:43", "content": "

Yeah, I thought about reconstructing 3D model of a car and background. As far as I know those methods have very high precision (in submillimeter range sometimes).

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "208556": "Another way to approach the problem is multi-view segmentation or 3d reconstruction. Anyone considering methods like these:\n\n - https://hal.inria.fr/inria-00568915/document\n\n - http://igl.ethz.ch/projects/light-field-segmentation/\n\n - http://www.oliverwang.info/\n (Efficient 3D Object Segmentation from Densely Sampled Light Fields with Applications to 3D Reconstruction)\n\n - https://www.youtube.com/watch?v=UPS8ZPHPY5A\n\n - http://chenlab.ece.cornell.edu/projects/iModel/\n\n https://www.youtube.com/watch?v=iPsutuvbWbc\n\n - http://www.cv-foundation.org/openaccess/content_iccv_2013/papers/Djelouah_Multi-view_Object_Segmentation_2013_ICCV_paper.pdf\n\n - http://adjelouah.name/index.php/projects/multi-view-segmentation/\n\n - https://www.youtube.com/watch?v=UhwayReG9i8\n\n - https://www.youtube.com/watch?v=VoEBJfI8kDw\n\n - https://www.youtube.com/watch?v=462hKb86CIY\n\n - https://hal.inria.fr/hal-00735718/PDF/Final_N-tuple_Multi-View_Silhouette_Extraction-1.pdf\n\n - https://www.youtube.com/watch?v=LEi5qllwMKg\n\n- https://www.youtube.com/watch?v=skFFsJZ3xCM\n\n- https://www.youtube.com/watch?v=tjMpgno6k5A\n\n- https://www.youtube.com/watch?v=PrUVo3potl4\n\n- http://www.apsipa.org/proceedings_2014/Data/paper/1267.pdf\n\n\nof course it is possible to develop from these ideas and make a deep CNN for it (e.g. using CNN to replace some prediction or model function)", "208576": "Yeah, I thought about reconstructing 3D model of a car and background. As far as I know those methods have very high precision (in submillimeter range sometimes)." }, "source": "meta" }