← All publications
VidSeg: Zero-Shot Video Semantic Segmentation based on Pre-Trained Diffusion Models
Abstract
Repurposes pre-trained diffusion models for zero-shot semantic segmentation of video, removing the need for task-specific labeled training data.