# Multiresolution Tree Networks for 3D Point Cloud Processing

**Authors:** Matheus Gadelha, Rui Wang, Subhransu Maji  
**arXiv:** [1807.03520](https://arxiv.org/abs/1807.03520)  
**Date:** 2018-07-10  
**Summary page:** https://www.summarizepaper.com/en/paper/1807.03520v2/

## Key points

- "Multiresolution Tree Networks for 3D Point Cloud Processing" introduces a novel approach to process point clouds efficiently
- The proposed multiresolution tree-structured networks represent a 3D shape as locality-preserving 1D ordered lists of points at multiple resolutions
- Allows for efficient feed-forward processing through 1D convolutions and enables coarse-to-fine analysis through a multi-grid architecture
- Results in faster convergence and smaller memory footprint during training
- Demonstrates superior performance compared to existing methods in shape classification, image-to-shape inference, and unsupervised learning tasks
- Offers an innovative solution for processing point clouds in 3D shape understanding and generation tasks
- Potential applications in computer vision and graphics

## Summary

, , , , 

The paper titled "Multiresolution Tree Networks for 3D Point Cloud Processing" introduces a novel approach to efficiently process point clouds for various 3D shape understanding and generation tasks. The proposed multiresolution tree-structured networks represent a 3D shape as locality-preserving 1D ordered lists of points at multiple resolutions, allowing for efficient feed-forward processing through 1D convolutions and enabling coarse-to-fine analysis through a multi-grid architecture. This results in faster convergence and a smaller memory footprint during training. The authors demonstrate the effectiveness of their approach by applying it to shape classification, image-to-shape inference, and unsupervised learning tasks, achieving superior performance compared to existing methods. In conclusion, the multiresolution tree networks offer an innovative solution for processing point clouds in 3D shape understanding and generation tasks, with potential applications in various fields such as computer vision and graphics.