LiLa-Net: Lightweight Latent LiDAR Autoencoder for 3D Point Cloud Reconstruction
This is an incremental improvement for autonomous vehicle perception systems, focusing on efficiency in resource-constrained settings.
The paper tackles 3D point cloud reconstruction from LiDAR data in traffic environments by proposing LiLa-Net, a lightweight autoencoder that uses simplified skip connections and fewer encoder layers, achieving improved reconstruction quality and strong generalization to unrelated objects.
This work proposed a 3D autoencoder architecture, named LiLa-Net, which encodes efficient features from real traffic environments, employing only the LiDAR's point clouds. For this purpose, we have real semi-autonomous vehicle, equipped with Velodyne LiDAR. The system leverage skip connections concept to improve the performance without using extensive resources as the state-of-the-art architectures. Key changes include reducing the number of encoder layers and simplifying the skip connections, while still producing an efficient and representative latent space which allows to accurately reconstruct the original point cloud. Furthermore, an effective balance has been achieved between the information carried by the skip connections and the latent encoding, leading to improved reconstruction quality without compromising performance. Finally, the model demonstrates strong generalization capabilities, successfully reconstructing objects unrelated to the original traffic environment.