Tabula rasa: model transfer for object category detection
Our objective is transfer training of a discriminatively trained object category detector, in order to reduce the number of training images required. To this end we propose three transfer learning formulations where a template learnt previously for other categories is used to regularize the training...
| Main Authors: | , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
IEEE
2012
|
| _version_ | 1824458672723460096 |
|---|---|
| author | Aytar, Y Zisserman, A |
| author_facet | Aytar, Y Zisserman, A |
| author_sort | Aytar, Y |
| collection | OXFORD |
| description | Our objective is transfer training of a discriminatively trained object category detector, in order to reduce the number of training images required. To this end we propose three transfer learning formulations where a template learnt previously for other categories is used to regularize the training of a new category. All the formulations result in convex optimization problems. Experiments (on PASCAL VOC) demonstrate significant performance gains by transfer learning from one class to another (e.g. motorbike to bicycle), including one-shot learning, specialization from class to a subordinate class (e.g. from quadruped to horse) and transfer using multiple components. In the case of multiple training samples it is shown that a detection performance approaching that of the state of the art can be achieved with substantially fewer training samples. |
| first_indexed | 2025-02-19T04:29:37Z |
| format | Conference item |
| id | oxford-uuid:3ad6d906-3b6f-464f-9faa-e4614cb05bf1 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2025-02-19T04:29:37Z |
| publishDate | 2012 |
| publisher | IEEE |
| record_format | dspace |
| spelling | oxford-uuid:3ad6d906-3b6f-464f-9faa-e4614cb05bf12024-12-17T11:20:31ZTabula rasa: model transfer for object category detectionConference itemhttp://purl.org/coar/resource_type/c_5794uuid:3ad6d906-3b6f-464f-9faa-e4614cb05bf1EnglishSymplectic ElementsIEEE2012Aytar, YZisserman, AOur objective is transfer training of a discriminatively trained object category detector, in order to reduce the number of training images required. To this end we propose three transfer learning formulations where a template learnt previously for other categories is used to regularize the training of a new category. All the formulations result in convex optimization problems. Experiments (on PASCAL VOC) demonstrate significant performance gains by transfer learning from one class to another (e.g. motorbike to bicycle), including one-shot learning, specialization from class to a subordinate class (e.g. from quadruped to horse) and transfer using multiple components. In the case of multiple training samples it is shown that a detection performance approaching that of the state of the art can be achieved with substantially fewer training samples. |
| spellingShingle | Aytar, Y Zisserman, A Tabula rasa: model transfer for object category detection |
| title | Tabula rasa: model transfer for object category detection |
| title_full | Tabula rasa: model transfer for object category detection |
| title_fullStr | Tabula rasa: model transfer for object category detection |
| title_full_unstemmed | Tabula rasa: model transfer for object category detection |
| title_short | Tabula rasa: model transfer for object category detection |
| title_sort | tabula rasa model transfer for object category detection |
| work_keys_str_mv | AT aytary tabularasamodeltransferforobjectcategorydetection AT zissermana tabularasamodeltransferforobjectcategorydetection |