FM-ChangeNet: Learning Change through Pathwise Feature Transport
2026-07-06 • Artificial Intelligence
Artificial IntelligenceComputer Vision and Pattern Recognition
AI summaryⓘ
The authors propose FM-ChangeNet, a new way to detect changes between two images taken at different times. Instead of just comparing the images at the start and end, their method looks at a smooth path of intermediate changes in a feature space, which helps the model understand how things evolve over time. They also learn a velocity field that shows how features move, which helps distinguish real changes from irrelevant differences like lighting or small misalignments. Their approach uses a multi-scale architecture and achieves strong results on remote sensing data.
change detectionfeature spacevelocity fieldbi-temporal reasoninglatent statesremote sensingsegmentation lossmulti-scale architecturespatial alignmenttrajectory consistency
Authors
Roie Kazoom, George Leifman, Genady Beryozkin
Abstract
We present FM-ChangeNet, a pathwise-supervised framework for change detection that reformulates bi-temporal reasoning as continuous transport in feature space rather than static endpoint comparison. Given encoded pre and post-temporal representations, we construct intermediate latent states and learn a time-conditioned velocity field $\hat{v}_θ(z_t,t)$ along the transformation trajectory. This pathwise formulation constrains the predictor over a continuum of intermediate states, providing a denser and less ambiguous supervision signal than conventional endpoint-only segmentation and enabling the model to capture temporal evolution explicitly. The learned velocity field is not only a transport mechanism but also an interpretable representation of change: its magnitude serves as a spatially localized change cue that helps distinguish true structural variation from nuisance effects such as illumination shifts and spatial misalignment. We develop a hierarchical multi-scale architecture with cross-temporal alignment, time-conditioned coarse-to-fine flow decoding, and a unified objective that couples flow supervision, trajectory consistency, spatial regularization, and segmentation loss. Experiments on remote sensing benchmarks show that the proposed framework produces more structured and robust change representations while achieving state-of-the-art performance.