Annotation-free Generation of Training Data Using Mixed Domains for Segmentation of 3D LiDAR Point Clouds
Semantic segmentation is important for robots navigating with 3D LiDARs, but the generation of training datasets requires tedious manual effort. In this paper, we introduce a set of strategies to efficiently generate large datasets by combining real and synthetic data samples. More specifically, the...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Sciendo
2025-09-01
|
| Series: | Foundations of Computing and Decision Sciences |
| Subjects: | |
| Online Access: | https://doi.org/10.2478/fcds-2025-0013 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849225044101693440 |
|---|---|
| author | Cop Konrad Sułek Bartosz Trzciński Tomasz |
| author_facet | Cop Konrad Sułek Bartosz Trzciński Tomasz |
| author_sort | Cop Konrad |
| collection | DOAJ |
| description | Semantic segmentation is important for robots navigating with 3D LiDARs, but the generation of training datasets requires tedious manual effort. In this paper, we introduce a set of strategies to efficiently generate large datasets by combining real and synthetic data samples. More specifically, the method populates recorded empty scenes with navigation-relevant obstacles generated synthetically, thus combining two domains: real life and synthetic. Our approach requires no manual annotation, no detailed knowledge about actual data feature distribution, and no real-life data of objects of interest. We validate the proposed method in the underground parking scenario and compare it with available open-source datasets. The experiments show superiority to the off-the-shelf datasets containing similar data characteristics but also highlight the difficulty of achieving the level of manually annotated datasets. We also show that combining generated and annotated data improves the performance visibly, especially for cases with rare occurrences of objects of interest. Our solution is suitable for direct application in robotic systems. |
| format | Article |
| id | doaj-art-a586bcb6bcd34bd6b8db38f4e89b598a |
| institution | Kabale University |
| issn | 2300-3405 |
| language | English |
| publishDate | 2025-09-01 |
| publisher | Sciendo |
| record_format | Article |
| series | Foundations of Computing and Decision Sciences |
| spelling | doaj-art-a586bcb6bcd34bd6b8db38f4e89b598a2025-08-25T06:11:49ZengSciendoFoundations of Computing and Decision Sciences2300-34052025-09-0150334737110.2478/fcds-2025-0013Annotation-free Generation of Training Data Using Mixed Domains for Segmentation of 3D LiDAR Point CloudsCop Konrad0Sułek Bartosz1Trzciński Tomasz21Warsaw University of Technology, Faculty of Electronics and Information TechnologyNowowiejska 15/19, 00-665Warsaw, Poland2United Robots Sp. z o.o., Świeradowska 47, 02-622Warsaw, Poland1Warsaw University of Technology, Faculty of Electronics and Information TechnologyNowowiejska 15/19, 00-665Warsaw, PolandSemantic segmentation is important for robots navigating with 3D LiDARs, but the generation of training datasets requires tedious manual effort. In this paper, we introduce a set of strategies to efficiently generate large datasets by combining real and synthetic data samples. More specifically, the method populates recorded empty scenes with navigation-relevant obstacles generated synthetically, thus combining two domains: real life and synthetic. Our approach requires no manual annotation, no detailed knowledge about actual data feature distribution, and no real-life data of objects of interest. We validate the proposed method in the underground parking scenario and compare it with available open-source datasets. The experiments show superiority to the off-the-shelf datasets containing similar data characteristics but also highlight the difficulty of achieving the level of manually annotated datasets. We also show that combining generated and annotated data improves the performance visibly, especially for cases with rare occurrences of objects of interest. Our solution is suitable for direct application in robotic systems.https://doi.org/10.2478/fcds-2025-0013deep learning for visual perceptionsemantic segmentation3d li-darrobotic perceptionpoint clouds |
| spellingShingle | Cop Konrad Sułek Bartosz Trzciński Tomasz Annotation-free Generation of Training Data Using Mixed Domains for Segmentation of 3D LiDAR Point Clouds Foundations of Computing and Decision Sciences deep learning for visual perception semantic segmentation 3d li-dar robotic perception point clouds |
| title | Annotation-free Generation of Training Data Using Mixed Domains for Segmentation of 3D LiDAR Point Clouds |
| title_full | Annotation-free Generation of Training Data Using Mixed Domains for Segmentation of 3D LiDAR Point Clouds |
| title_fullStr | Annotation-free Generation of Training Data Using Mixed Domains for Segmentation of 3D LiDAR Point Clouds |
| title_full_unstemmed | Annotation-free Generation of Training Data Using Mixed Domains for Segmentation of 3D LiDAR Point Clouds |
| title_short | Annotation-free Generation of Training Data Using Mixed Domains for Segmentation of 3D LiDAR Point Clouds |
| title_sort | annotation free generation of training data using mixed domains for segmentation of 3d lidar point clouds |
| topic | deep learning for visual perception semantic segmentation 3d li-dar robotic perception point clouds |
| url | https://doi.org/10.2478/fcds-2025-0013 |
| work_keys_str_mv | AT copkonrad annotationfreegenerationoftrainingdatausingmixeddomainsforsegmentationof3dlidarpointclouds AT sułekbartosz annotationfreegenerationoftrainingdatausingmixeddomainsforsegmentationof3dlidarpointclouds AT trzcinskitomasz annotationfreegenerationoftrainingdatausingmixeddomainsforsegmentationof3dlidarpointclouds |