Improved Training for 3D Point Cloud Classification

The point cloud is a 3D geometric data of irregular format. As a result, they are needed to be transformed into 3D voxels or a collection of images before being fed into models. This unnecessarily increases the volume of the data and the complexities of dealing with it. PointNet is a pioneering approach in this direction that feeds the 3D point cloud data directly to a model. This research work is developed on top of the existing PointNet architecture. The ModelNet10 dataset, a collection of 3D images with 10 class labels, has been used for this study. The goal of the study is to improve the accuracy of PointNet. To achieve this, a few variations of encoder models have been proposed along with improved training protocol, and transfer learning from larger datasets in this research work. Also, an extensive hyperparameter study has been done. The experiments in this research work achieve a 6.10% improvement over the baseline model. The code for this work is publicly available at



Results from the Paper

  Submit results from this paper to get state-of-the-art GitHub badges and help the community compare results to other papers.


No methods listed for this paper. Add relevant methods here