Reproduce “SeFi-CD: A Semantic First Change Detection Paradigm That Can Detect Any Change You Want” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ling Zhao, Zhenyang Huang, Dongsheng Kuang, Chengli Peng, Jun Gan, Haifeng Li
The AUWCD framework implements a "Semantic-First Change Detection" (SeFi-CD) paradigm, enabling users to detect any specified change in remote sensing images through natural language prompts without retraining. This approach leverages Vision-Language Models and Foundation Segment Models, demonstrating superior adaptability over traditional visual-first methods.
Key insights
- Change detection is fundamentally a semantic-driven task, and prioritizing user-defined semantic intent over general visual features leads to more intuitive and adaptable systems.
- The integration of pre-trained large multimodal foundation models (VLMs and FSMs) enables zero-shot or few-shot adaptation to diverse change detection tasks without specific per-task training data.
- Negative textual prompts for 'Change Regions of Uninterest' (CRoUI) can effectively suppress unwanted detections, enhancing the accuracy and focus on the desired semantic changes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.