Discovering relationships between object categories via universal canonical maps
We tackle the problem of learning the geometry of multiple categories of deformable objects jointly. Recent work has shown that it is possible to learn a unified dense pose predictor for several categories of related objects. However, training such models requires to initialize inter-category corres...
| Hlavní autoři: | , , , , |
|---|---|
| Médium: | Conference item |
| Jazyk: | English |
| Vydáno: |
IEEE
2021
|
| _version_ | 1826284042748690432 |
|---|---|
| author | Neverova, N Sanakoyeu, A Labatut, P Novotny, D Vedaldi, A |
| author_facet | Neverova, N Sanakoyeu, A Labatut, P Novotny, D Vedaldi, A |
| author_sort | Neverova, N |
| collection | OXFORD |
| description | We tackle the problem of learning the geometry of multiple categories of deformable objects jointly. Recent work has shown that it is possible to learn a unified dense pose predictor for several categories of related objects. However, training such models requires to initialize inter-category correspondences by hand. This is suboptimal and the resulting models fail to maintain correct correspondences as individual categories are learned. In this paper, we show that improved correspondences can be learned automatically as a natural byproduct of learning category-specific dense pose predictors. To do this, we express correspondences between different categories and between images and categories using a unified embedding. Then, we use the latter to enforce two constraints: symmetric inter-category cycle consistency and a new asymmetric image-to-category cycle consistency. Without any manual annotations for the intercategory correspondences, we obtain state-of-the-art alignment results, outperforming dedicated methods for matching 3D shapes. Moreover, the new model is also better at the task of dense pose prediction than prior work. |
| first_indexed | 2024-03-07T01:07:55Z |
| format | Conference item |
| id | oxford-uuid:8bfc6594-01db-460d-b70f-d208e67d3c6d |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T01:07:55Z |
| publishDate | 2021 |
| publisher | IEEE |
| record_format | dspace |
| spelling | oxford-uuid:8bfc6594-01db-460d-b70f-d208e67d3c6d2022-03-26T22:41:43ZDiscovering relationships between object categories via universal canonical mapsConference itemhttp://purl.org/coar/resource_type/c_5794uuid:8bfc6594-01db-460d-b70f-d208e67d3c6dEnglishSymplectic ElementsIEEE2021Neverova, NSanakoyeu, ALabatut, PNovotny, DVedaldi, AWe tackle the problem of learning the geometry of multiple categories of deformable objects jointly. Recent work has shown that it is possible to learn a unified dense pose predictor for several categories of related objects. However, training such models requires to initialize inter-category correspondences by hand. This is suboptimal and the resulting models fail to maintain correct correspondences as individual categories are learned. In this paper, we show that improved correspondences can be learned automatically as a natural byproduct of learning category-specific dense pose predictors. To do this, we express correspondences between different categories and between images and categories using a unified embedding. Then, we use the latter to enforce two constraints: symmetric inter-category cycle consistency and a new asymmetric image-to-category cycle consistency. Without any manual annotations for the intercategory correspondences, we obtain state-of-the-art alignment results, outperforming dedicated methods for matching 3D shapes. Moreover, the new model is also better at the task of dense pose prediction than prior work. |
| spellingShingle | Neverova, N Sanakoyeu, A Labatut, P Novotny, D Vedaldi, A Discovering relationships between object categories via universal canonical maps |
| title | Discovering relationships between object categories via universal canonical maps |
| title_full | Discovering relationships between object categories via universal canonical maps |
| title_fullStr | Discovering relationships between object categories via universal canonical maps |
| title_full_unstemmed | Discovering relationships between object categories via universal canonical maps |
| title_short | Discovering relationships between object categories via universal canonical maps |
| title_sort | discovering relationships between object categories via universal canonical maps |
| work_keys_str_mv | AT neverovan discoveringrelationshipsbetweenobjectcategoriesviauniversalcanonicalmaps AT sanakoyeua discoveringrelationshipsbetweenobjectcategoriesviauniversalcanonicalmaps AT labatutp discoveringrelationshipsbetweenobjectcategoriesviauniversalcanonicalmaps AT novotnyd discoveringrelationshipsbetweenobjectcategoriesviauniversalcanonicalmaps AT vedaldia discoveringrelationshipsbetweenobjectcategoriesviauniversalcanonicalmaps |