Search papers, labs, and topics across Lattice.
The paper introduces MaSoN (Make Some Noise), an unsupervised change detection (UCD) framework for remote sensing that synthesizes changes directly in the latent feature space during training, avoiding reliance on predefined assumptions about change types. MaSoN dynamically estimates changes using feature statistics of target data, enabling data-driven variation aligned with the target domain and facilitating generalization across diverse change types. The method achieves state-of-the-art performance on five benchmarks, improving the average F1 score by 14.1 percentage points compared to existing UCD methods.
By synthesizing changes directly in the latent space, MaSoN achieves state-of-the-art unsupervised change detection in remote sensing, generalizing to complex and rare scenarios where existing methods fall short.
Unsupervised change detection (UCD) in remote sensing aims to localise semantic changes between two images of the same region without relying on labelled data during training. Most recent approaches rely either on frozen foundation models in a training-free manner or on training with synthetic changes generated in pixel space. Both strategies inherently rely on predefined assumptions about change types, typically introduced through handcrafted rules, external datasets, or auxiliary generative models. Due to these assumptions, such methods fail to generalise beyond a few change types, limiting their real-world usage, especially in rare or complex scenarios. To address this, we propose MaSoN (Make Some Noise), an end-to-end UCD framework that synthesises diverse changes directly in the latent feature space during training. It generates changes that are dynamically estimated using feature statistics of target data, enabling diverse yet data-driven variation aligned with the target domain. It also easily extends to new modalities, such as SAR. MaSoN generalises strongly across diverse change types and achieves state-of-the-art performance on five benchmarks, improving the average F1 score by 14.1 percentage points. Project page: https://blaz-r.github.io/mason_ucd
