Abstract
| Original language | English |
|---|---|
| Title of host publication | 2019 IEEE Intelligent Vehicles Symposium, IV 2019 |
| Place of Publication | Piscataway |
| Publisher | Institute of Electrical and Electronics Engineers |
| Pages | 1334-1339 |
| Number of pages | 6 |
| ISBN (Electronic) | 978-1-7281-0560-4 |
| DOIs | |
| Publication status | Published - 2019 |
| Event | 2019 IEEE Intelligent Vehicles Symposium (IV) - Paris, France Duration: 9 Jun 2019 → 12 Jun 2019 |
Conference
| Conference | 2019 IEEE Intelligent Vehicles Symposium (IV) |
|---|---|
| Country | France |
| City | Paris |
| Period | 9/06/19 → 12/06/19 |
Fingerprint
Cite this
}
On boosting semantic street scene segmentation with weak supervision. / Meletis, Panagiotis; Dubbelman, Gijs.
2019 IEEE Intelligent Vehicles Symposium, IV 2019. Piscataway : Institute of Electrical and Electronics Engineers, 2019. p. 1334-1339.Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review
TY - GEN
T1 - On boosting semantic street scene segmentation with weak supervision
AU - Meletis, Panagiotis
AU - Dubbelman, Gijs
PY - 2019
Y1 - 2019
N2 - Training convolutional networks for semantic segmentation requires per-pixel ground truth labels, which are very time consuming and hence costly to obtain. Therefore, in this work, we research and develop a hierarchical deep network architecture and the corresponding loss for semantic segmentation that can be trained from weak supervision, such as bounding boxes or image level labels, as well as from strong per-pixel supervision. We demonstrate that the hierarchical structure and the simultaneous training on strong (per-pixel) and weak (bounding boxes) labels, even from separate datasets, constantly increases the performance against per-pixel only training. Moreover, we explore the more challenging case of adding weak image-level labels. We collect street scene images and weak labels from the immense Open Images dataset to generate the OpenScapes dataset, and we use this novel dataset to increase segmentation performance on two established per-pixel labeled datasets, Cityscapes and Vistas. We report performance gains up to +13.2% mIoU on crucial street scene classes, and inference speed of 20 fps on a Titan V GPU for Cityscapes at 512 x 1024 resolution. Our network and OpenScapes dataset are shared with the research community.
AB - Training convolutional networks for semantic segmentation requires per-pixel ground truth labels, which are very time consuming and hence costly to obtain. Therefore, in this work, we research and develop a hierarchical deep network architecture and the corresponding loss for semantic segmentation that can be trained from weak supervision, such as bounding boxes or image level labels, as well as from strong per-pixel supervision. We demonstrate that the hierarchical structure and the simultaneous training on strong (per-pixel) and weak (bounding boxes) labels, even from separate datasets, constantly increases the performance against per-pixel only training. Moreover, we explore the more challenging case of adding weak image-level labels. We collect street scene images and weak labels from the immense Open Images dataset to generate the OpenScapes dataset, and we use this novel dataset to increase segmentation performance on two established per-pixel labeled datasets, Cityscapes and Vistas. We report performance gains up to +13.2% mIoU on crucial street scene classes, and inference speed of 20 fps on a Titan V GPU for Cityscapes at 512 x 1024 resolution. Our network and OpenScapes dataset are shared with the research community.
U2 - 10.1109/IVS.2019.8814217
DO - 10.1109/IVS.2019.8814217
M3 - Conference contribution
SP - 1334
EP - 1339
BT - 2019 IEEE Intelligent Vehicles Symposium, IV 2019
PB - Institute of Electrical and Electronics Engineers
CY - Piscataway
ER -