Scene Level Label

Scene-level labeling focuses on assigning semantic labels to entire 3D scenes or images, rather than individual objects or points, to reduce the need for extensive manual annotation. Current research emphasizes leveraging limited labeled data through techniques like pseudo-labeling, incorporating information from multiple modalities (e.g., RGB-D images and point clouds), and employing advanced architectures such as implicit neural fields and large language models to improve segmentation accuracy. This approach is significant because it significantly reduces the cost and effort associated with data annotation, enabling broader application of 3D scene understanding in fields like remote sensing, autonomous driving, and robotics.