Loading…
Dyna-DM: Dynamic object-aware self-supervised monocular depth maps
Self-supervised monocular depth estimation has been a subject of intense study in recent years, because of its applications in robotics and autonomous driving. Much of the recent work focuses on improving depth estimation by increasing architecture complexity. This paper shows that state-of-the-art...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Default Conference proceeding |
| Published: |
2023
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/2134/26484766.v1 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1818164618393026560 |
|---|---|
| author | Kieran Saunders George Vogiatzis Luis J Manso |
| author_facet | Kieran Saunders George Vogiatzis Luis J Manso |
| author_sort | Kieran Saunders (17247865) |
| collection | Figshare |
| description | Self-supervised monocular depth estimation has been a subject of intense study in recent years, because of its applications in robotics and autonomous driving. Much of the recent work focuses on improving depth estimation by increasing architecture complexity. This paper shows that state-of-the-art performance can also be achieved by improving the learning process rather than increasing model complexity. More specifically, we propose (i) disregarding small potentially dynamic objects when training, and (ii) employing an appearance-based approach to separately estimate object pose for truly dynamic objects. We demonstrate that these simplifications reduce GPU memory usage by 29% and result in qualitatively and quantitatively improved depth maps. |
| format | Default Conference proceeding |
| id | rr-article-26484766 |
| institution | Loughborough University |
| publishDate | 2023 |
| record_format | Figshare |
| spelling | rr-article-264847662023-05-25T00:00:00Z Dyna-DM: Dynamic object-aware self-supervised monocular depth maps Kieran Saunders (17247865) George Vogiatzis (14326170) Luis J Manso (19320370) Training Computer vision Image analysis Self-supervised monocular depth estimation has been a subject of intense study in recent years, because of its applications in robotics and autonomous driving. Much of the recent work focuses on improving depth estimation by increasing architecture complexity. This paper shows that state-of-the-art performance can also be achieved by improving the learning process rather than increasing model complexity. More specifically, we propose (i) disregarding small potentially dynamic objects when training, and (ii) employing an appearance-based approach to separately estimate object pose for truly dynamic objects. We demonstrate that these simplifications reduce GPU memory usage by 29% and result in qualitatively and quantitatively improved depth maps.<p></p> 2023-05-25T00:00:00Z Text Conference contribution 2134/26484766.v1 https://figshare.com/articles/conference_contribution/Dyna-DM_Dynamic_object-aware_self-supervised_monocular_depth_maps/26484766 All Rights Reserved |
| spellingShingle | Training Computer vision Image analysis Kieran Saunders George Vogiatzis Luis J Manso Dyna-DM: Dynamic object-aware self-supervised monocular depth maps |
| title | Dyna-DM: Dynamic object-aware self-supervised monocular depth maps |
| title_full | Dyna-DM: Dynamic object-aware self-supervised monocular depth maps |
| title_fullStr | Dyna-DM: Dynamic object-aware self-supervised monocular depth maps |
| title_full_unstemmed | Dyna-DM: Dynamic object-aware self-supervised monocular depth maps |
| title_short | Dyna-DM: Dynamic object-aware self-supervised monocular depth maps |
| title_sort | dyna-dm: dynamic object-aware self-supervised monocular depth maps |
| topic | Training Computer vision Image analysis |
| url | https://hdl.handle.net/2134/26484766.v1 |