Submitted by Rongliang Cheng.

Submission data

Full name3DASPP for Semantic Label and Semantic Context Embedding.
DescriptionFirstly, we used a 3D ASPP module for the semantic segmentation of 3d point cloud.
Secondly, we embedded the semantic context feature to global scene classification, thus can enhance the semantic correlation explicitly.
In addition, we used softmax to enhance the learning for semantic context feature during training stage.
Input Data TypesUses Geometry        Uses 3D
Programming language(s)Python with CUDA
HardwareNvidia V100
Submission creation date17 Jun, 2021
Last edited17 Jun, 2021

Scene type classification results



Infoavg iouapartmentbathroombedroom / hotelbookstore / libraryconference roomcopy/mail roomhallwaykitchenlaundry roomliving room / loungemiscofficestorage / basement / garage
0.5560.5000.9380.7780.6671.0000.2500.5000.7500.3330.5000.0000.8120.200