UniFunc3D: Unified Active Spatial-Temporal Grounding for 3D Affordance Segmentation
UniFunc3D introduces a unified framework using a single multimodal large language model to segment functional objects in 3D scenes from video and text instructions. It uses an active spatial-temporal grounding process with a coarse-to-fine strategy across multiple frames to jointly perform semantic, temporal, and spatial reasoning without needing task-specific training.