RESSCAL3D: A Resolution-Scalable Deep Learning Approach for 3D Semantic Segmentation of Point Clouds
RESSCAL3D is a novel deep learning-based architecture that enables resolution-scalable 3D semantic segmentation of point clouds, allowing early decision-making and efficient processing of additional points as they become available.
Geometrically-Driven Aggregation for Enhancing Zero-Shot 3D Point Cloud Understanding
Geometrically-driven aggregation of vision-language model representations can effectively improve the quality of zero-shot 3D point cloud understanding across various downstream tasks.
Probability-Driven Framework for Open World 3D Point Cloud Semantic Segmentation
A novel probability-driven framework (PDF) that leverages probability outputs to identify unknown objects and incrementally expand the knowledge base for open world 3D point cloud semantic segmentation.
Learnable Earth Parser: Unsupervised Decomposition of Large-Scale Aerial 3D Scans into Interpretable Prototypes
An unsupervised deep learning method that decomposes large-scale aerial 3D point clouds into a small set of learned prototypical 3D shapes, enabling interpretable reconstruction, semantic segmentation, and instance segmentation of complex real-world scenes.
