Point Transformer, Zhao, Jiang, Jia, Torr, Koltun; 2020 - Summary
author: kelseyball
score: 10 / 10

What is the core idea?

How is it realized (technically)?

Point Transformer Layer

Position Encoding Function

Point Transformer Block

Point Transformer Network

Training details

How well does the paper perform?

Point Transformers achieve new state-of-the-art on semantic segmentation, shape classification, and object part segmentation, outperforming a variety of models including pointwise MLPs, voxel-based models, graph-based models, sparse convolutional networks, and continuous convolutional networks.

Semantic Segmentation

Shape Classification

Object Part Segmentation

What interesting variants are explored?

Takeaways from ablation study: