PCGen: A Fully Parallelizable Point Cloud Generative Model.

Generative models have the potential to revolutionize 3D extended reality. A primary obstacle is that augmented and virtual reality need real-time computing. Current state-of-the-art point cloud random generation methods are not fast enough for these applications. We introduce a vector-quantized variational autoencoder model (VQVAE) that can synthesize high-quality point clouds in milliseconds. Unlike previous work in VQVAEs, our model offers a compact sample representation suitable for conditional generation and data exploration with potential applications in rapid prototyping. We achieve this result by combining architectural improvements with an innovative approach for probabilistic random generation. First, we rethink current parallel point cloud autoencoder structures, and we propose several solutions to improve robustness, efficiency and reconstruction quality. Notable contributions in the decoder architecture include an innovative computation layer to process the shape semantic information, an attention mechanism that helps the model focus on different areas and a filter to cover possible sampling errors. Secondly, we introduce a parallel sampling strategy for VQVAE models consisting of a double encoding system, where a variational autoencoder learns how to generate the complex discrete distribution of the VQVAE, not only allowing quick inference but also describing the shape with a few global variables. We compare the proposed decoder and our VQVAE model with established and concurrent work, and we prove, one by one, the validity of the single contributions.

GPU Rasterization-Based 3D LiDAR Simulation for Deep Learning.

High-quality data are of utmost importance for any deep-learning application. However, acquiring such data and their annotation is challenging. This paper presents a GPU-accelerated simulator that enables the generation of high-quality, perfectly labelled data for any Time-of-Flight sensor, including LiDAR. Our approach optimally exploits the 3D graphics pipeline of the GPU, significantly decreasing data generation time while preserving compatibility with all real-time rendering engines. The presented algorithms are generic and allow users to perfectly mimic the unique sampling pattern of any such sensor. To validate our simulator, two neural networks are trained for denoising and semantic segmentation. To bridge the gap between reality and simulation, a novel loss function is introduced that requires only a small set of partially annotated real data. It enables the learning of classes for which no labels are provided in the real data, hence dramatically reducing annotation efforts. With this work, we hope to provide means for alleviating the data acquisition problem that is pertinent to deep-learning applications.