NeuralSampler: Euclidean Point Cloud Auto-Encoder and Sampler
This addresses a limitation in deep learning for 3D point cloud generation, offering a more flexible and efficient solution for applications like computer graphics and robotics.
The paper tackles the problem of generating 3D point clouds with fixed sizes and large network parameters by proposing an auto-encoder that handles arbitrary sizes and upsamples sparse clouds, achieving better performance with less than half the parameters of state-of-the-art methods.
Most algorithms that rely on deep learning-based approaches to generate 3D point sets can only produce clouds containing fixed number of points. Furthermore, they typically require large networks parameterized by many weights, which makes them hard to train. In this paper, we propose an auto-encoder architecture that can both encode and decode clouds of arbitrary size and demonstrate its effectiveness at upsampling sparse point clouds. Interestingly, we can do so using less than half as many parameters as state-of-the-art architectures while still delivering better performance. We will make our code base fully available.