ChangeFlow -- Latent Rectified Flow for Change Detection in Remote Sensing
ChangeFlow is a new generative framework designed for change detection in remote sensing images. It addresses the limitations of existing methods by synthesizing change masks in latent space, improving robustness and confidence estimation. The framework achieves an average F1 score of 80.4%, surpassing previous methods while maintaining competitive inference speed.
- ▪ChangeFlow reformulates change detection as the synthesis of a change mask in latent space via rectified flow.
- ▪The framework uses a structured yet lightweight conditioning signal to guide the change detection process.
- ▪ChangeFlow improves the average F1 score by 1.3 points over the previous best method across four benchmarks.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computer Vision and Pattern Recognition arXiv:2605.15375 (cs) [Submitted on 14 May 2026] Title:ChangeFlow -- Latent Rectified Flow for Change Detection in Remote Sensing Authors:Blaž Rolih, Matic Fučka, Filip Wolf, Luka Čehovin Zajc View a PDF of the paper titled ChangeFlow -- Latent Rectified Flow for Change Detection in Remote Sensing, by Bla\v{z} Rolih and 3 other authors View PDF HTML (experimental) Abstract:Remote sensing change detection (RSCD) aims to localise changes between two images of the same geographic region. In practice, change masks often follow region-level annotation conventions rather than purely local appearance differences, making them context-dependent and occasionally ambiguous.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
