TOWARDS ACCURATE INSTANCE SEGMENTATION IN LARGE-SCALE LIDAR POINT CLOUDS
Panoptic segmentation is the combination of semantic and instance segmentation: assign the points in a 3D point cloud to semantic categories <em>and</em> partition them into distinct object instances. It has many obvious applications for outdoor scene understanding, from city mapping to...
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2023-12-01
|
| Series: | ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| Online Access: | https://isprs-annals.copernicus.org/articles/X-1-W1-2023/605/2023/isprs-annals-X-1-W1-2023-605-2023.pdf |
| _version_ | 1827594771429851136 |
|---|---|
| author | B. Xiang T. Peters T. Kontogianni F. Vetterli S. Puliti R. Astrup K. Schindler |
| author_facet | B. Xiang T. Peters T. Kontogianni F. Vetterli S. Puliti R. Astrup K. Schindler |
| author_sort | B. Xiang |
| collection | DOAJ |
| description | Panoptic segmentation is the combination of semantic and instance segmentation: assign the points in a 3D point cloud to semantic categories <em>and</em> partition them into distinct object instances. It has many obvious applications for outdoor scene understanding, from city mapping to forest management. Existing methods struggle to segment nearby instances of the same semantic category, like adjacent pieces of street furniture or neighbouring trees, which limits their usability for inventory- or management-type applications that rely on object instances. This study explores the steps of the panoptic segmentation pipeline concerned with clustering points into object instances, with the goal to alleviate that bottleneck. We find that a carefully designed clustering strategy, which leverages multiple types of learned point embeddings, significantly improves instance segmentation. Experiments on the <em>NPM3D</em> urban mobile mapping dataset and the <em>FOR-instance</em> forest dataset demonstrate the effectiveness and versatility of the proposed strategy. |
| first_indexed | 2024-03-09T02:41:09Z |
| format | Article |
| id | doaj.art-36dc7c9d907045b898c3ad7299681142 |
| institution | Directory Open Access Journal |
| issn | 2194-9042 2194-9050 |
| language | English |
| last_indexed | 2024-03-09T02:41:09Z |
| publishDate | 2023-12-01 |
| publisher | Copernicus Publications |
| record_format | Article |
| series | ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| spelling | doaj.art-36dc7c9d907045b898c3ad72996811422023-12-06T03:10:10ZengCopernicus PublicationsISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences2194-90422194-90502023-12-01X-1-W1-202360561210.5194/isprs-annals-X-1-W1-2023-605-2023TOWARDS ACCURATE INSTANCE SEGMENTATION IN LARGE-SCALE LIDAR POINT CLOUDSB. Xiang0T. Peters1T. Kontogianni2F. Vetterli3S. Puliti4R. Astrup5K. Schindler6Institute of Geodesy and Photogrammetry, ETH Zürich, Zürich, SwitzerlandInstitute of Geodesy and Photogrammetry, ETH Zürich, Zürich, SwitzerlandInstitute of Geodesy and Photogrammetry, ETH Zürich, Zürich, SwitzerlandInstitute of Geodesy and Photogrammetry, ETH Zürich, Zürich, SwitzerlandNorwegian Institute of Bioeconomy Research (NIBIO), NorwayNorwegian Institute of Bioeconomy Research (NIBIO), NorwayInstitute of Geodesy and Photogrammetry, ETH Zürich, Zürich, SwitzerlandPanoptic segmentation is the combination of semantic and instance segmentation: assign the points in a 3D point cloud to semantic categories <em>and</em> partition them into distinct object instances. It has many obvious applications for outdoor scene understanding, from city mapping to forest management. Existing methods struggle to segment nearby instances of the same semantic category, like adjacent pieces of street furniture or neighbouring trees, which limits their usability for inventory- or management-type applications that rely on object instances. This study explores the steps of the panoptic segmentation pipeline concerned with clustering points into object instances, with the goal to alleviate that bottleneck. We find that a carefully designed clustering strategy, which leverages multiple types of learned point embeddings, significantly improves instance segmentation. Experiments on the <em>NPM3D</em> urban mobile mapping dataset and the <em>FOR-instance</em> forest dataset demonstrate the effectiveness and versatility of the proposed strategy.https://isprs-annals.copernicus.org/articles/X-1-W1-2023/605/2023/isprs-annals-X-1-W1-2023-605-2023.pdf |
| spellingShingle | B. Xiang T. Peters T. Kontogianni F. Vetterli S. Puliti R. Astrup K. Schindler TOWARDS ACCURATE INSTANCE SEGMENTATION IN LARGE-SCALE LIDAR POINT CLOUDS ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| title | TOWARDS ACCURATE INSTANCE SEGMENTATION IN LARGE-SCALE LIDAR POINT CLOUDS |
| title_full | TOWARDS ACCURATE INSTANCE SEGMENTATION IN LARGE-SCALE LIDAR POINT CLOUDS |
| title_fullStr | TOWARDS ACCURATE INSTANCE SEGMENTATION IN LARGE-SCALE LIDAR POINT CLOUDS |
| title_full_unstemmed | TOWARDS ACCURATE INSTANCE SEGMENTATION IN LARGE-SCALE LIDAR POINT CLOUDS |
| title_short | TOWARDS ACCURATE INSTANCE SEGMENTATION IN LARGE-SCALE LIDAR POINT CLOUDS |
| title_sort | towards accurate instance segmentation in large scale lidar point clouds |
| url | https://isprs-annals.copernicus.org/articles/X-1-W1-2023/605/2023/isprs-annals-X-1-W1-2023-605-2023.pdf |
| work_keys_str_mv | AT bxiang towardsaccurateinstancesegmentationinlargescalelidarpointclouds AT tpeters towardsaccurateinstancesegmentationinlargescalelidarpointclouds AT tkontogianni towardsaccurateinstancesegmentationinlargescalelidarpointclouds AT fvetterli towardsaccurateinstancesegmentationinlargescalelidarpointclouds AT spuliti towardsaccurateinstancesegmentationinlargescalelidarpointclouds AT rastrup towardsaccurateinstancesegmentationinlargescalelidarpointclouds AT kschindler towardsaccurateinstancesegmentationinlargescalelidarpointclouds |