3DGNN for RGB-D segmentation
This is the Pytorch implementation of 3D Graph Neural Networks for RGBD Semantic Segmentation:
Data Preparation
- Download NYU_Depth_V2 dataset from here and select scenes and save as
./datasets/data/nyu_depth_v2_labeled.mat - Transfer depth images to hha by yourself from here and save in
./datasets/data/hha/.
Emviroment
Required CUDA (8.0) + pytorch 0.4.1