The paper introduces MaSoN, an end-to-end unsupervised remote sensing change detection framework that synthesises diverse changes directly in latent feature space during training. By generating changes based on feature statistics of the target data, MaSoN produces data‑driven variations that align with the target domain and can be applied to new modalities such as SAR and multispectral imagery. The method achieves a 14.1 percentage point improvement in average F1 score across five benchmarks, demonstrating strong generalisation across diverse change types.
By Bla\v{z} Rolih, Matic Fu\v{c}ka, Filip Wolf, Luka \v{C}ehovin Zajc
The paper proposes a weakly supervised remote sensing change detection method that uses change captions as the sole supervision signal, eliminating the need for pixel‑level change masks. It introduces a caption‑driven generation pipeline to create bi‑temporal image pairs with controlled changes and a Semantic‑Appearance Agreement Framework (SAAF) that fuses caption‑grounded semantic responses with RGB differences for accurate change localization. Experiments on the Flair‑RSGen and WHU‑CDC datasets demonstrate that SAAF outperforms existing limited‑supervision baselines in macro‑averaged IoU and F1 metrics.
By Yuan Qian, Jie Ma
The paper introduces MEDIC, a framework for change captioning that explicitly models different change types such as color shifts and object additions. MEDIC uses type‑specialized memory experts that dynamically retrieve relevant visual patterns and softly route inputs across these experts, enabling each to focus on the most informative cues for its change category. Experiments show that MEDIC consistently outperforms existing methods on diverse and challenging datasets.
By Jiyoung Park, InJae Oh, Jung Uk Kim
HyperFlow is a method for class‑incremental semantic change detection in remote sensing, allowing a model to progressively learn new types of land‑cover changes. It leverages the hierarchical nature of land‑cover semantics by embedding categories in hyperbolic space and uses Busemann prototypes to model change as an anomaly in a learned semantic flow. The authors also introduce HiChange, a large‑scale dataset with fine‑grained hierarchical labels, and show that incremental refinement of categories improves both change detection and segmentation performance.
arXiv:2604. 20623v2 Announce Type: replace-cross Abstract: Traditional change detection identifies where changes occur, but does not explain what changed in natural language.
By Roie Kazoom, Yotam Gigi, George Leifman, Tomer Shekel, Genady Beryozkin
arXiv:2606. 28724v1 Announce Type: cross Abstract: Understanding and localizing subtle changes between paired images is critical for tasks such as surveillance and image editing.
By Jinhong Hu, Xiaoping Wang, Shuyin Huang, Guojin Zhong, Kaitai Liu, Kai Lu