GAGCN: Generative Adversarial Graph Convolutional Network for 3D Point Cloud Semantic Segmentation
Key Contributions:
- Proposed GCN with adversarial learning scheme in 3D point cloud segmentation
- Utilized Embedding loss for adversarial learning
- Proposed an effective way for 3D point cloud convolution
Instance Average IoU
PointNet++ | DGCNN | PointCNN | KPConv | Proposed witout Adv | Proposed with Adv |
---|---|---|---|---|---|
85.1 | 85.2 | 86.1 | 86.4 | 86.2 | 86.9 |
Class Average IoU
PointNet++ | DGCNN | PointCNN | KPConv | Proposed witout Adv | Proposed with Adv |
---|---|---|---|---|---|
80.4 | 82.3 | 84.6 | 85.1 | 84.1 | 85.9 |
This code has been tested on:
ubuntu
torch == 1.7.1
torch-geometric == 1.7.0
torch-cluster == 1.5.9
torch-scatter == 2.0.6
torch-sparse == 0.6.9
torch-spline-conv == 1.2.1
open3d == 0.9.0