Open world entity segmentation
<p>We introduce a new image segmentation task, called Entity Segmentation (ES), which aims to segment all visual entities (objects and stuffs) in an image without predicting their semantic labels. By removing the need of class label prediction, the models trained for such task can focus more o...
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Journal article |
| Sprog: | English |
| Udgivet: |
IEEE
2022
|
| _version_ | 1826311663204171776 |
|---|---|
| author | Qi, L Kuen, J Wang, Y Gu, J Zhao, H Torr, P Lin, Z Jia, J |
| author_facet | Qi, L Kuen, J Wang, Y Gu, J Zhao, H Torr, P Lin, Z Jia, J |
| author_sort | Qi, L |
| collection | OXFORD |
| description | <p>We introduce a new image segmentation task, called Entity Segmentation (ES), which aims to segment all visual entities (objects and stuffs) in an image without predicting their semantic labels. By removing the need of class label prediction, the models trained for such task can focus more on improving segmentation quality. It has many practical applications such as image manipulation and editing where the quality of segmentation masks is crucial but class labels are less important. We conduct the first-ever study to investigate the feasibility of convolutional center-based representation to segment things and stuffs in a unified manner, and show that such representation fits exceptionally well in the context of ES. More specifically, we propose a CondInst-like fully-convolutional architecture with two novel modules specifically designed to exploit the class-agnostic and non-overlapping requirements of ES. Experiments show that the models designed and trained for ES significantly outperforms popular class-specific panoptic segmentation models in terms of segmentation quality. Moreover, an ES model can be easily trained on a combination of multiple datasets without the need to resolve label conflicts in dataset merging, and the model trained for ES on one or more datasets can generalize very well to other test datasets of unseen domains. The code has been released at https://github.com/dvlab-research/Entity</p> |
| first_indexed | 2024-03-07T08:13:05Z |
| format | Journal article |
| id | oxford-uuid:badf7dab-2df6-4ccf-9bce-e9b870cf84ee |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T08:13:05Z |
| publishDate | 2022 |
| publisher | IEEE |
| record_format | dspace |
| spelling | oxford-uuid:badf7dab-2df6-4ccf-9bce-e9b870cf84ee2023-12-08T10:42:18ZOpen world entity segmentationJournal articlehttp://purl.org/coar/resource_type/c_dcae04bcuuid:badf7dab-2df6-4ccf-9bce-e9b870cf84eeEnglishSymplectic ElementsIEEE2022Qi, LKuen, JWang, YGu, JZhao, HTorr, PLin, ZJia, J<p>We introduce a new image segmentation task, called Entity Segmentation (ES), which aims to segment all visual entities (objects and stuffs) in an image without predicting their semantic labels. By removing the need of class label prediction, the models trained for such task can focus more on improving segmentation quality. It has many practical applications such as image manipulation and editing where the quality of segmentation masks is crucial but class labels are less important. We conduct the first-ever study to investigate the feasibility of convolutional center-based representation to segment things and stuffs in a unified manner, and show that such representation fits exceptionally well in the context of ES. More specifically, we propose a CondInst-like fully-convolutional architecture with two novel modules specifically designed to exploit the class-agnostic and non-overlapping requirements of ES. Experiments show that the models designed and trained for ES significantly outperforms popular class-specific panoptic segmentation models in terms of segmentation quality. Moreover, an ES model can be easily trained on a combination of multiple datasets without the need to resolve label conflicts in dataset merging, and the model trained for ES on one or more datasets can generalize very well to other test datasets of unseen domains. The code has been released at https://github.com/dvlab-research/Entity</p> |
| spellingShingle | Qi, L Kuen, J Wang, Y Gu, J Zhao, H Torr, P Lin, Z Jia, J Open world entity segmentation |
| title | Open world entity segmentation |
| title_full | Open world entity segmentation |
| title_fullStr | Open world entity segmentation |
| title_full_unstemmed | Open world entity segmentation |
| title_short | Open world entity segmentation |
| title_sort | open world entity segmentation |
| work_keys_str_mv | AT qil openworldentitysegmentation AT kuenj openworldentitysegmentation AT wangy openworldentitysegmentation AT guj openworldentitysegmentation AT zhaoh openworldentitysegmentation AT torrp openworldentitysegmentation AT linz openworldentitysegmentation AT jiaj openworldentitysegmentation |