2、Deep Learning in Higher Dimensions

Multi-Dimensional ConvNets

1D Convolutions

WaveNet


3D Classification

3D Semantic Segmentation

Volumetric Grids

3D Shape Completion on Grids

ScanNet(三维语义分割)

Semantic Segmentation in 3D

sliding window

SurfaceNet（立体重建）

Stereo Reconstruction

ScanComplete（完全卷积）

Fully Convolutional

Dependent Predictions（自回归神经网络）

Autoregressive Neural Networks

Spatial Extent

Coarse-to-Fine Predictions

conclusion so far

Volumetric Hierarchies

Discriminative Tasks

Generative Tasks

Conclusion so far

Multi-view

Classification

Segmentation

Hybrid: Volumetric + Multi-view

2D + 3D Semantic Segmentation

3D Volumetric + Multi-view





Conclusion so far

Point Clouds

PointNet

DeepLearning on Point Clouds

PointNet++

Point Convolutions

Conclusions so far

Point Sets(global)

Mesh-based

Conclusion so far

Sparse Convolutions

Conclusions so far

3D Scene Understanding