C3DPO: Canonical 3D pose networks for non-rigid structure from motion
We propose C3DPO, a method for extracting 3D models of deformable objects from 2D keypoint annotations in unconstrained images. We do so by learning a deep network that reconstructs a 3D object from a single view at a time, accounting for partial occlusions, and explicitly factoring the effects of v...
| Main Authors: | , , , , |
|---|---|
| Format: | Conference item |
| Published: |
2020
|
| _version_ | 1826270527072763904 |
|---|---|
| author | Novotny, D Ravi, N Graham, B Neverova, N Vedaldi, A |
| author_facet | Novotny, D Ravi, N Graham, B Neverova, N Vedaldi, A |
| author_sort | Novotny, D |
| collection | OXFORD |
| description | We propose C3DPO, a method for extracting 3D models of deformable objects from 2D keypoint annotations in unconstrained images. We do so by learning a deep network that reconstructs a 3D object from a single view at a time, accounting for partial occlusions, and explicitly factoring the effects of viewpoint changes and object deformations. In order to achieve this factorization, we introduce a novel regularization technique. We first show that the factorization is successful if, and only if, there exists a certain canonicalization function of the reconstructed shapes. Then, we learn the canonicalization function together with the reconstruction one, which constrains the result to be consistent. We demonstrate state-of-the-art reconstruction results for methods that do not use ground-truth 3D supervision for a number of benchmarks, including Up3D and PASCAL3D+. |
| first_indexed | 2024-03-06T21:42:12Z |
| format | Conference item |
| id | oxford-uuid:4851bf46-998f-4d20-9f9a-dc57c7bf29ca |
| institution | University of Oxford |
| last_indexed | 2024-03-06T21:42:12Z |
| publishDate | 2020 |
| record_format | dspace |
| spelling | oxford-uuid:4851bf46-998f-4d20-9f9a-dc57c7bf29ca2022-03-26T15:25:01ZC3DPO: Canonical 3D pose networks for non-rigid structure from motionConference itemhttp://purl.org/coar/resource_type/c_5794uuid:4851bf46-998f-4d20-9f9a-dc57c7bf29caSymplectic Elements at Oxford2020Novotny, DRavi, NGraham, BNeverova, NVedaldi, AWe propose C3DPO, a method for extracting 3D models of deformable objects from 2D keypoint annotations in unconstrained images. We do so by learning a deep network that reconstructs a 3D object from a single view at a time, accounting for partial occlusions, and explicitly factoring the effects of viewpoint changes and object deformations. In order to achieve this factorization, we introduce a novel regularization technique. We first show that the factorization is successful if, and only if, there exists a certain canonicalization function of the reconstructed shapes. Then, we learn the canonicalization function together with the reconstruction one, which constrains the result to be consistent. We demonstrate state-of-the-art reconstruction results for methods that do not use ground-truth 3D supervision for a number of benchmarks, including Up3D and PASCAL3D+. |
| spellingShingle | Novotny, D Ravi, N Graham, B Neverova, N Vedaldi, A C3DPO: Canonical 3D pose networks for non-rigid structure from motion |
| title | C3DPO: Canonical 3D pose networks for non-rigid structure from motion |
| title_full | C3DPO: Canonical 3D pose networks for non-rigid structure from motion |
| title_fullStr | C3DPO: Canonical 3D pose networks for non-rigid structure from motion |
| title_full_unstemmed | C3DPO: Canonical 3D pose networks for non-rigid structure from motion |
| title_short | C3DPO: Canonical 3D pose networks for non-rigid structure from motion |
| title_sort | c3dpo canonical 3d pose networks for non rigid structure from motion |
| work_keys_str_mv | AT novotnyd c3dpocanonical3dposenetworksfornonrigidstructurefrommotion AT ravin c3dpocanonical3dposenetworksfornonrigidstructurefrommotion AT grahamb c3dpocanonical3dposenetworksfornonrigidstructurefrommotion AT neverovan c3dpocanonical3dposenetworksfornonrigidstructurefrommotion AT vedaldia c3dpocanonical3dposenetworksfornonrigidstructurefrommotion |