PointNet解析
返回总结列表 https://blog.****.net/albert199199/article/details/103531486
PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation
Key points:
-
Point-wise
-
坐标作为channel（N*1的图像卷积，卷积核1*1）
-
通过max pooling实现对称函数，但会造成信息损失
-
T-net目标：实现旋转不变性
-
计算单元：1*1卷积，MLP
-
实验：3D classification (ModelNet40, sampling from mesh), object part segmentation(ShapeNet part dataset) and semantic scene sgementation (Stanford 3D semantic parsing dataset)
-
code