Voxel Representation
Voxel representations are three-dimensional grid-based structures used to encode and process 3D data, aiming to balance computational efficiency with detailed geometric information. Current research focuses on improving the accuracy and efficiency of voxel-based models for various applications, including 3D object detection, scene reconstruction, and medical image analysis, often employing deep neural networks, attention mechanisms, and novel fusion techniques to integrate voxel data with other modalities like point clouds and images. These advancements are significantly impacting fields like autonomous driving, robotics, and medical imaging by enabling faster and more accurate 3D scene understanding and manipulation.
Papers
Modeling Uncertainty in 3D Gaussian Splatting through Continuous Semantic Splatting
Joey Wilson, Marcelino Almeida, Min Sun, Sachit Mahajan, Maani Ghaffari, Parker Ewen, Omid Ghasemalizadeh, Cheng-Hao Kuo, Arnie Sen
GVKF: Gaussian Voxel Kernel Functions for Highly Efficient Surface Reconstruction in Open Scenes
Gaochao Song, Chong Cheng, Hao Wang