Discriminate-and-Rectify Encoders: Learning from Image Transformation Sets
The complexity of a learning task is increased by transformations in the input space that preserve class identity. Visual object recognition for example is affected by changes in viewpoint, scale, illumination or planar transformations. While drastically altering the visual appearance, these changes...
| Main Authors: | , , |
|---|---|
| Format: | Technical Report |
| Language: | en_US |
| Published: |
Center for Brains, Minds and Machines (CBMM), arXiv
2017
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/1721.1/107446 |
| _version_ | 1811088706023456768 |
|---|---|
| author | Tachetti, Andrea Voinea, Stephen Evangelopoulos, Georgios |
| author_facet | Tachetti, Andrea Voinea, Stephen Evangelopoulos, Georgios |
| author_sort | Tachetti, Andrea |
| collection | MIT |
| description | The complexity of a learning task is increased by transformations in the input space that preserve class identity. Visual object recognition for example is affected by changes in viewpoint, scale, illumination or planar transformations. While drastically altering the visual appearance, these changes are orthogonal to recognition and should not be reflected in the representation or feature encoding used for learning. We introduce a framework for weakly supervised learning of image embeddings that are robust to transformations and selective to the class distribution, using sets of transforming examples (orbit sets), deep parametrizations and a novel orbit-based loss. The proposed loss combines a discriminative, contrastive part for orbits with a reconstruction error that learns to rectify orbit transformations. The learned embeddings are evaluated in distance metric-based tasks, such as one-shot classification under geometric transformations, as well as face verification and retrieval under more realistic visual variability. Our results suggest that orbit sets, suitably computed or observed, can be used for efficient, weakly-supervised learning of semantically relevant image embeddings. |
| first_indexed | 2024-09-23T14:06:15Z |
| format | Technical Report |
| id | mit-1721.1/107446 |
| institution | Massachusetts Institute of Technology |
| language | en_US |
| last_indexed | 2024-09-23T14:06:15Z |
| publishDate | 2017 |
| publisher | Center for Brains, Minds and Machines (CBMM), arXiv |
| record_format | dspace |
| spelling | mit-1721.1/1074462019-04-11T13:40:28Z Discriminate-and-Rectify Encoders: Learning from Image Transformation Sets Tachetti, Andrea Voinea, Stephen Evangelopoulos, Georgios supervised learning object recognition machine learning The complexity of a learning task is increased by transformations in the input space that preserve class identity. Visual object recognition for example is affected by changes in viewpoint, scale, illumination or planar transformations. While drastically altering the visual appearance, these changes are orthogonal to recognition and should not be reflected in the representation or feature encoding used for learning. We introduce a framework for weakly supervised learning of image embeddings that are robust to transformations and selective to the class distribution, using sets of transforming examples (orbit sets), deep parametrizations and a novel orbit-based loss. The proposed loss combines a discriminative, contrastive part for orbits with a reconstruction error that learns to rectify orbit transformations. The learned embeddings are evaluated in distance metric-based tasks, such as one-shot classification under geometric transformations, as well as face verification and retrieval under more realistic visual variability. Our results suggest that orbit sets, suitably computed or observed, can be used for efficient, weakly-supervised learning of semantically relevant image embeddings. This material is based upon work supported by the Center for Brains, Minds and Machines (CBMM), funded by NSF STC award CCF-1231216. 2017-03-16T19:50:32Z 2017-03-16T19:50:32Z 2017-03-13 Technical Report Working Paper Other http://hdl.handle.net/1721.1/107446 arXiv:1703.04775v1 en_US CBMM Memo Series;062 Attribution-NonCommercial-ShareAlike 3.0 United States http://creativecommons.org/licenses/by-nc-sa/3.0/us/ application/pdf Center for Brains, Minds and Machines (CBMM), arXiv |
| spellingShingle | supervised learning object recognition machine learning Tachetti, Andrea Voinea, Stephen Evangelopoulos, Georgios Discriminate-and-Rectify Encoders: Learning from Image Transformation Sets |
| title | Discriminate-and-Rectify Encoders: Learning from Image Transformation Sets |
| title_full | Discriminate-and-Rectify Encoders: Learning from Image Transformation Sets |
| title_fullStr | Discriminate-and-Rectify Encoders: Learning from Image Transformation Sets |
| title_full_unstemmed | Discriminate-and-Rectify Encoders: Learning from Image Transformation Sets |
| title_short | Discriminate-and-Rectify Encoders: Learning from Image Transformation Sets |
| title_sort | discriminate and rectify encoders learning from image transformation sets |
| topic | supervised learning object recognition machine learning |
| url | http://hdl.handle.net/1721.1/107446 |
| work_keys_str_mv | AT tachettiandrea discriminateandrectifyencoderslearningfromimagetransformationsets AT voineastephen discriminateandrectifyencoderslearningfromimagetransformationsets AT evangelopoulosgeorgios discriminateandrectifyencoderslearningfromimagetransformationsets |