FM-ChangeNet: Learning Change through Pathwise Feature Transport

2026-07-06

AI Summary

This work addresses the limitations of traditional bi-temporal change detection methods, which rely on static endpoint comparisons and are thus vulnerable to illumination variations and registration errors, while failing to model temporal evolution. The authors reformulate the task as a continuous path transport problem in feature space, explicitly modeling intermediate latent state trajectories by learning a time-conditioned velocity field. A novel path supervision mechanism is introduced, leveraging the magnitude of the velocity field to provide interpretable local change cues and incorporating dense intermediate-state supervision to enhance discriminative robustness and structural consistency. The proposed model employs a multi-scale hierarchical architecture that integrates cross-temporal alignment, coarse-to-fine optical flow decoding, and a unified objective function combining optical flow supervision, trajectory consistency, spatial regularization, and segmentation loss. It achieves state-of-the-art performance on remote sensing change detection benchmarks, yielding more structured and robust change representations against common disturbances.

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.

Problem

• change detection
• bi-temporal reasoning
• feature transport
• remote sensing
• temporal evolution

Innovation

• pathwise feature transport
• time-conditioned velocity field
• continuous trajectory
• change detection
• multi-scale architecture