Paper reading group for SRIP 2019

Goal

We will focus on 3D reconstruction and point cloud paper over 2019 summer. The group member is mainly from drone group, but students from other groups are also welcomed. Hopefully, we can have a basic understanding of the progress in those 2 fields. The weekly meeting is designed for paper discussion and idea brainstorming. There is no need to make slides or treat this event as anything formal. Come by with your coffee and throw out some crazy ideas ^_^. Feel free to edit this doc as well.

Resource

• https://github.com/Yochengliu/awesome-point-cloud-analysis
• https://github.com/timzhang642/3D-Machine-Learning

Time

Tuesday 2 - 3 pm

Location

CSE grad lounge

Wishlist

You are more than welcomed to propose papers related to this topic, but not covered in our meeting. List your wishlist below.

• Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications
• Deep Kd-Networks for the Recognition of 3D Point Cloud Models
• What Do Single-view 3D Reconstruction Networks Learn?
• Volumetric and Multi-View CNNs for Object Classification on 3D Data
• A Point Set Generation Network for 3D Object Reconstruction from a Single Image
• Octree Generating Networks: Efficient Convolutional Architectures for High-resolution 3D Outputs
• Joint Embeddings of Shapes and Images via CNN Image Purification

Schedule

Week1

• PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation
• PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space
• 3DMatch: Learning Local Geometric Descriptors from RGB-D Reconstructions
• ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes
• Scan2CAD: Learning CAD Model Alignment in RGB-D Scans

Week2

• SO-Net: Self-Organizing Network for Point Cloud Analysis
• PU-Net: Point Cloud Upsampling Network
• 3D Semantic Segmentation with Submanifold Sparse Convolutional Networks
• SPLATNet: Sparse Lattice Networks for Point Cloud Processing
• PointCNN: Convolution On X-Transformed Points
• Learning Dense Correspondence via 3D-guided Cycle Consistency

Week3

• Learning Efficient Point Cloud Generation for Dense 3D Object Reconstruction
• Deep Reinforcement Learning of Volume-guided Progressive View Inpainting for 3D Point Scene Completion from a Single Depth Image
• 3D Point Scene Completion from a Single Depth Image
• PointConv: Deep Convolutional Networks on 3D Point Clouds
• Spatial Transformer for 3D Points
• Unpaired Point Cloud Completion on Real Scans using Adversarial Training
• Zero-shot Learning of 3D Point Cloud Objects

Week4 (Mesh)

• Neural 3D Mesh Renderer
• Pixel2Mesh
• Learning Category-Specific Mesh Reconstruction from Image Collections
• Mesh R-CNN

Week5 (Primitives)

• Learning Shape Abstractions by Assembling Volumetric Primitives
• Learning Shape Templates with Structured Implicit Functions
• Learning View Priors for Single-view 3D Reconstruction
• 3D-PRNN: Generating Shape Primitives with Recurrent Neural Networks
• Grass: Generative recursive autoencoders for shape structures

Participants

Sean Kamano, Po Hsiang Huang, Dacheng Li, Jayi Wang, Yixuan Huang, Brandon Leung, Chih-Hui Ho, Amir Persekian