AssemblyNetV2: A Dataset for Predicting Exploded Views Through Point Clouds

AssemblyNetV2 contains two datasets: the original AssemblyNet dataset, resampled with face normals, and a novel dataset for predicting whether parts can be disassembled, also containing both points and normals. The original AssemblyNet dataset contains 5420 samples while the new dataset contains 16337 samples of 512-point and 1024-point point clouds of mixed industrial assemblies and LEGO models. The new dataset for part removal contains annotations for binary classification of whether a part can be removed from an assembly. The intended use case of the two datasets is to predict exploded views of assemblies by iteratively removing parts in the predicted directions.