Poster: Zero-Shot Prompt-Based Partial 3D Point Cloud Creation from an Unlabeled 2D Image

This work explores a novel zero-shot framework to generate partial 3D point clouds from a single unlabeled 2D image. By combining prompt-based object detection, segmentation, and depth estimation, our approach creates highquality 3D representations with minimal manual intervention. The generated models hold promise for augmenting training datasets with synthetic data of underrepresented objects for applications such as autonomous driving.