Learning to reconstruct shapes from unseen classes
© 2018 Curran Associates Inc.All rights reserved. From a single image, humans are able to perceive the full 3D shape of an object by exploiting learned shape priors from everyday life. Contemporary single-image 3D reconstruction algorithms aim to solve this task in a similar fashion, but often end u...
| Main Authors: | , , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
2021
|
| Online Access: | https://hdl.handle.net/1721.1/137406 |
| _version_ | 1826188868627464192 |
|---|---|
| author | Zhang, Xiuming Zhang, Zhoutong Zhang, Chengkai Tenenbaum, Joshua B. Freeman, William T. Wu, Jiajun |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Zhang, Xiuming Zhang, Zhoutong Zhang, Chengkai Tenenbaum, Joshua B. Freeman, William T. Wu, Jiajun |
| author_sort | Zhang, Xiuming |
| collection | MIT |
| description | © 2018 Curran Associates Inc.All rights reserved. From a single image, humans are able to perceive the full 3D shape of an object by exploiting learned shape priors from everyday life. Contemporary single-image 3D reconstruction algorithms aim to solve this task in a similar fashion, but often end up with priors that are highly biased by training classes. Here we present an algorithm, Generalizable Reconstruction (GenRe), designed to capture more generic, class-agnostic shape priors. We achieve this with an inference network and training procedure that combine 2.5D representations of visible surfaces (depth and silhouette), spherical shape representations of both visible and non-visible surfaces, and 3D voxel-based representations, in a principled manner that exploits the causal structure of how 3D shapes give rise to 2D images. Experiments demonstrate that GenRe performs well on single-view shape reconstruction, and generalizes to diverse novel objects from categories not seen during training. |
| first_indexed | 2024-09-23T08:06:16Z |
| format | Article |
| id | mit-1721.1/137406 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T08:06:16Z |
| publishDate | 2021 |
| record_format | dspace |
| spelling | mit-1721.1/1374062022-09-30T07:33:00Z Learning to reconstruct shapes from unseen classes Zhang, Xiuming Zhang, Zhoutong Zhang, Chengkai Tenenbaum, Joshua B. Freeman, William T. Wu, Jiajun Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory © 2018 Curran Associates Inc.All rights reserved. From a single image, humans are able to perceive the full 3D shape of an object by exploiting learned shape priors from everyday life. Contemporary single-image 3D reconstruction algorithms aim to solve this task in a similar fashion, but often end up with priors that are highly biased by training classes. Here we present an algorithm, Generalizable Reconstruction (GenRe), designed to capture more generic, class-agnostic shape priors. We achieve this with an inference network and training procedure that combine 2.5D representations of visible surfaces (depth and silhouette), spherical shape representations of both visible and non-visible surfaces, and 3D voxel-based representations, in a principled manner that exploits the causal structure of how 3D shapes give rise to 2D images. Experiments demonstrate that GenRe performs well on single-view shape reconstruction, and generalizes to diverse novel objects from categories not seen during training. 2021-11-04T19:32:26Z 2021-11-04T19:32:26Z 2018 2019-05-28T12:34:12Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/137406 Zhang, Xiuming, Zhang, Zhoutong, Zhang, Chengkai, Tenenbaum, Joshua B., Freeman, William T. et al. 2018. "Learning to reconstruct shapes from unseen classes." en https://papers.nips.cc/paper/7494-learning-to-reconstruct-shapes-from-unseen-classes Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. application/pdf Neural Information Processing Systems (NIPS) |
| spellingShingle | Zhang, Xiuming Zhang, Zhoutong Zhang, Chengkai Tenenbaum, Joshua B. Freeman, William T. Wu, Jiajun Learning to reconstruct shapes from unseen classes |
| title | Learning to reconstruct shapes from unseen classes |
| title_full | Learning to reconstruct shapes from unseen classes |
| title_fullStr | Learning to reconstruct shapes from unseen classes |
| title_full_unstemmed | Learning to reconstruct shapes from unseen classes |
| title_short | Learning to reconstruct shapes from unseen classes |
| title_sort | learning to reconstruct shapes from unseen classes |
| url | https://hdl.handle.net/1721.1/137406 |
| work_keys_str_mv | AT zhangxiuming learningtoreconstructshapesfromunseenclasses AT zhangzhoutong learningtoreconstructshapesfromunseenclasses AT zhangchengkai learningtoreconstructshapesfromunseenclasses AT tenenbaumjoshuab learningtoreconstructshapesfromunseenclasses AT freemanwilliamt learningtoreconstructshapesfromunseenclasses AT wujiajun learningtoreconstructshapesfromunseenclasses |