The field of 3D point cloud processing and Gaussian Splatting is rapidly evolving, with a focus on improving domain generalization, robustness, and security. Recent developments have explored the use of category-level geometry learning, transferable class statistics, and multi-scale feature approximation to enhance 3D object detection and segmentation. Additionally, researchers have investigated the use of multimodal data, such as Near-Infrared imagery and textual metadata, to improve 3D reconstruction in challenging environments like agriculture. The security of 3D point cloud models has also been a topic of interest, with the development of transfer-based black-box attack methods and evaluations of semantic residuals after object removal. Notable papers in this area include:

• Domain-aware Category-level Geometry Learning Segmentation for 3D Point Clouds, which proposes a category-level geometry learning framework for domain generalized 3D semantic segmentation.
• Remove360, which introduces a novel benchmark and evaluation framework to measure semantic residuals after object removal in 3D Gaussian Splatting.
• ComplicitSplat, which presents a black-box attack that exploits standard 3DGS shading methods to create viewpoint-specific camouflage.
• GALA, which proposes a novel framework for open-vocabulary 3D scene understanding with 3D Gaussian Splatting.
• Reconstruction Using the Invisible, which introduces a novel multimodal dataset and architecture for enhanced 3D Gaussian Splatting in agricultural scenes.
• Towards a 3D Transfer-based Black-box Attack, which proposes a novel transfer-based black-box attack method that improves the transferability of adversarial point clouds via critical feature guidance.