OmniOVCD: A New Framework for Open-Vocabulary Change Detection Using SAM 3

A research paper on arXiv introduces OmniOVCD, a standalone framework for Open-Vocabulary Change Detection (OVCD) in remote sensing. The framework leverages the Segment Anything Model 3 (SAM 3), which integrates segmentation and identification within a single promptable model. Existing training-free OVCD methods often rely on CLIP for category identification and additional models like DINO for feature extraction, leading to feature mismatching and instability. OmniOVCD proposes a Synergistic Fusion to Instance Decoupling (SFID) strategy that fuses semantic information from SAM 3's decoupled output heads. This approach streamlines the OVCD pipeline by eliminating the need for multiple models, improving stability and performance. The paper is published on arXiv with ID 2601.13895.