Learning to discover novel visual categories via deep transfer clustering
We consider the problem of discovering novel object categories in an image collection. While these images are unlabelled, we also assume prior knowledge of related but different image classes. We use such prior knowledge to reduce the ambiguity of clustering, and improve the quality of the newly dis...
| Main Authors: | , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
IEEE
2020
|
| _version_ | 1797087248112418816 |
|---|---|
| author | Han, K Vedaldi, A Zisserman, A |
| author_facet | Han, K Vedaldi, A Zisserman, A |
| author_sort | Han, K |
| collection | OXFORD |
| description | We consider the problem of discovering novel object categories in an image collection. While these images are unlabelled, we also assume prior knowledge of related but different image classes. We use such prior knowledge to reduce the ambiguity of clustering, and improve the quality of the newly discovered classes. Our contributions are twofold. The first contribution is to extend Deep Embedded Clustering to a transfer learning setting; we also improve the algorithm by introducing a representation bottleneck, temporal ensembling, and consistency. The second contribution is a method to estimate the number of classes in the unlabelled data. This also transfers knowledge from the known classes, using them as probes to diagnose different choices for the number of classes in the unlabelled subset. We thoroughly evaluate our method, substantially outperforming state-of-the-art techniques in a large number of benchmarks, including ImageNet, OmniGlot, CIFAR-100, CIFAR-10, and SVHN. |
| first_indexed | 2024-03-07T02:33:07Z |
| format | Conference item |
| id | oxford-uuid:a7e7bc2a-d3cb-457e-906a-8001012725ae |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T02:33:07Z |
| publishDate | 2020 |
| publisher | IEEE |
| record_format | dspace |
| spelling | oxford-uuid:a7e7bc2a-d3cb-457e-906a-8001012725ae2022-03-27T02:57:38ZLearning to discover novel visual categories via deep transfer clusteringConference itemhttp://purl.org/coar/resource_type/c_5794uuid:a7e7bc2a-d3cb-457e-906a-8001012725aeEnglishSymplectic Elements at OxfordIEEE2020Han, KVedaldi, AZisserman, AWe consider the problem of discovering novel object categories in an image collection. While these images are unlabelled, we also assume prior knowledge of related but different image classes. We use such prior knowledge to reduce the ambiguity of clustering, and improve the quality of the newly discovered classes. Our contributions are twofold. The first contribution is to extend Deep Embedded Clustering to a transfer learning setting; we also improve the algorithm by introducing a representation bottleneck, temporal ensembling, and consistency. The second contribution is a method to estimate the number of classes in the unlabelled data. This also transfers knowledge from the known classes, using them as probes to diagnose different choices for the number of classes in the unlabelled subset. We thoroughly evaluate our method, substantially outperforming state-of-the-art techniques in a large number of benchmarks, including ImageNet, OmniGlot, CIFAR-100, CIFAR-10, and SVHN. |
| spellingShingle | Han, K Vedaldi, A Zisserman, A Learning to discover novel visual categories via deep transfer clustering |
| title | Learning to discover novel visual categories via deep transfer clustering |
| title_full | Learning to discover novel visual categories via deep transfer clustering |
| title_fullStr | Learning to discover novel visual categories via deep transfer clustering |
| title_full_unstemmed | Learning to discover novel visual categories via deep transfer clustering |
| title_short | Learning to discover novel visual categories via deep transfer clustering |
| title_sort | learning to discover novel visual categories via deep transfer clustering |
| work_keys_str_mv | AT hank learningtodiscovernovelvisualcategoriesviadeeptransferclustering AT vedaldia learningtodiscovernovelvisualcategoriesviadeeptransferclustering AT zissermana learningtodiscovernovelvisualcategoriesviadeeptransferclustering |