MultiPartAssembly Benchmark V0.2.0 Release

The official release for the Breaking Bad Dataset benchmark (NeurIPS 2022 Datasets and Benchmarks Track)

MultiPartAssembly Benchmark V0.1.0 Release

Highlights

• Support two assembly datasets: PartNet (semantic assembly) and Breaking Bad (geometric assembly)
• Support two baseline assembly methods: Global and LSTM
• Support two state-of-the-art assembly methods: DGL and RGL-Net
• All the algorithms match their performance reported in the papers
• We try to adopt Transformer models for the part assembly task and achieve competitive results with state-of-the-art