3D Semantic Segmentation
The GOOSE 3D Semantic Segmentation Challenge was a competition hosted in conjunction with the Workshop on Field Robotics at ICRA 2025 in Atlanta, USA. The challenge is based on the annotated point cloud data of unstructured outdoor environments found in the GOOSE and GOOSE-Ex datasets.
Participants were tasked with developing and testing models for 3D Semantic Segmentation of LiDAR point clouds taken from three different robotic platforms. The teams with the best-performing models were awarded prizes and the opportunity to present their approach during the poster session of the Workshop on Field Robotics at ICRA 2025.
The classes are grouped into 9 superclasses for the purpose of this competition.
The superclasses are
other ,
artifical_structures ,
artificial_ground ,
natural_ground ,
obstacle ,
vehicle ,
vegetation ,
human ,
sky :
category_name,label_key,hex
other,0,#A9A9A9
artificial_structures,1,#DE88DE
artificial_ground,2,#EBFF3B
natural_ground,3,#A1887F
obstacle,4,#FFC107
vehicle,5,#F44336
vegetation,6,#4CAF50
human,7,#8FB0FF
sky,8,#2196F3
This simplifies the semantic segmentation task while still requiring the recognition of relevant surface and object types.
2025 Finalists
- 🥇 Xiaoya Zhang from 株式会社EARTHBRAIN with a mIoU of 82.0% · [technical report]