Nested Invariance Pooling and RBM Hashing for Image Instance Retrieval
The goal of this work is the computation of very compact binary hashes for image instance retrieval. Our approach has two novel contributions. The first one is Nested Invariance Pooling (NIP), a method inspired from i-theory, a mathematical theory for computing group invariant transformations with f...
| Main Authors: | , , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Published: |
Association for Computing Machinery (ACM)
2017
|
| Online Access: | http://hdl.handle.net/1721.1/112288 https://orcid.org/0000-0002-3944-0455 |
| _version_ | 1811073606754500608 |
|---|---|
| author | Morère, Olivier Lin, Jie Veillard, Antoine Duan, Ling-Yu Chandrasekhar, Vijay Poggio, Tomaso A |
| author2 | McGovern Institute for Brain Research at MIT. Center for Brains, Minds, and Machines |
| author_facet | McGovern Institute for Brain Research at MIT. Center for Brains, Minds, and Machines Morère, Olivier Lin, Jie Veillard, Antoine Duan, Ling-Yu Chandrasekhar, Vijay Poggio, Tomaso A |
| author_sort | Morère, Olivier |
| collection | MIT |
| description | The goal of this work is the computation of very compact binary hashes for image instance retrieval. Our approach has two novel contributions. The first one is Nested Invariance Pooling (NIP), a method inspired from i-theory, a mathematical theory for computing group invariant transformations with feed-forward neural networks. NIP is able to produce compact and well-performing descriptors with visual representations extracted from convolutional neural networks. We specifically incorporate scale, translation and rotation invariances but the scheme can be extended to any arbitrary sets of transformations. We also show that using moments of increasing order throughout nesting is important. The NIP descriptors are then hashed to the target code size (32-256 bits) with a Restricted Boltzmann Machine with a novel batch-level reg-ularization scheme specifically designed for the purpose of hashing (RBMH). A thorough empirical evaluation with state-of-the-art shows that the results obtained both with the NIP descriptors and the NIP+RBMH hashes are consistently outstanding across a wide range of datasets. |
| first_indexed | 2024-09-23T09:35:38Z |
| format | Article |
| id | mit-1721.1/112288 |
| institution | Massachusetts Institute of Technology |
| last_indexed | 2024-09-23T09:35:38Z |
| publishDate | 2017 |
| publisher | Association for Computing Machinery (ACM) |
| record_format | dspace |
| spelling | mit-1721.1/1122882022-09-30T15:33:29Z Nested Invariance Pooling and RBM Hashing for Image Instance Retrieval Morère, Olivier Lin, Jie Veillard, Antoine Duan, Ling-Yu Chandrasekhar, Vijay Poggio, Tomaso A McGovern Institute for Brain Research at MIT. Center for Brains, Minds, and Machines Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences Massachusetts Institute of Technology. Laboratory for Computational and Statistical Learning McGovern Institute for Brain Research at MIT Poggio, Tomaso A The goal of this work is the computation of very compact binary hashes for image instance retrieval. Our approach has two novel contributions. The first one is Nested Invariance Pooling (NIP), a method inspired from i-theory, a mathematical theory for computing group invariant transformations with feed-forward neural networks. NIP is able to produce compact and well-performing descriptors with visual representations extracted from convolutional neural networks. We specifically incorporate scale, translation and rotation invariances but the scheme can be extended to any arbitrary sets of transformations. We also show that using moments of increasing order throughout nesting is important. The NIP descriptors are then hashed to the target code size (32-256 bits) with a Restricted Boltzmann Machine with a novel batch-level reg-ularization scheme specifically designed for the purpose of hashing (RBMH). A thorough empirical evaluation with state-of-the-art shows that the results obtained both with the NIP descriptors and the NIP+RBMH hashes are consistently outstanding across a wide range of datasets. 2017-11-27T16:09:39Z 2017-11-27T16:09:39Z 2017-06 2017-11-16T18:39:36Z Article http://purl.org/eprint/type/ConferencePaper 978-1-4503-4701-3 http://hdl.handle.net/1721.1/112288 Morère, Olivier et al. “Nested Invariance Pooling and RBM Hashing for Image Instance Retrieval.” Proceedings of the 2017 ACM on International Conference on Multimedia Retrieval (ICMR ’17), June 6-9 2017, Bucharest, Romania, Association for Computing Machinery (ACM), June 2017 © 2017 Association for Computing Machinery (ACM) https://orcid.org/0000-0002-3944-0455 http://dx.doi.org/10.1145/3078971.3078987 Proceedings of the 2017 ACM on International Conference on Multimedia Retrieval (ICMR '17) Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf Association for Computing Machinery (ACM) arXiv |
| spellingShingle | Morère, Olivier Lin, Jie Veillard, Antoine Duan, Ling-Yu Chandrasekhar, Vijay Poggio, Tomaso A Nested Invariance Pooling and RBM Hashing for Image Instance Retrieval |
| title | Nested Invariance Pooling and RBM Hashing for Image Instance Retrieval |
| title_full | Nested Invariance Pooling and RBM Hashing for Image Instance Retrieval |
| title_fullStr | Nested Invariance Pooling and RBM Hashing for Image Instance Retrieval |
| title_full_unstemmed | Nested Invariance Pooling and RBM Hashing for Image Instance Retrieval |
| title_short | Nested Invariance Pooling and RBM Hashing for Image Instance Retrieval |
| title_sort | nested invariance pooling and rbm hashing for image instance retrieval |
| url | http://hdl.handle.net/1721.1/112288 https://orcid.org/0000-0002-3944-0455 |
| work_keys_str_mv | AT morereolivier nestedinvariancepoolingandrbmhashingforimageinstanceretrieval AT linjie nestedinvariancepoolingandrbmhashingforimageinstanceretrieval AT veillardantoine nestedinvariancepoolingandrbmhashingforimageinstanceretrieval AT duanlingyu nestedinvariancepoolingandrbmhashingforimageinstanceretrieval AT chandrasekharvijay nestedinvariancepoolingandrbmhashingforimageinstanceretrieval AT poggiotomasoa nestedinvariancepoolingandrbmhashingforimageinstanceretrieval |