![[2602.19881] Make Some Noise: Unsupervised Remote Sensing Change Detection Using Latent Space Perturbations](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19881] Make Some Noise: Unsupervised Remote Sensing Change Detection Using Latent Space Perturbations
Summary
The paper presents MaSoN, an innovative framework for unsupervised change detection in remote sensing that generates diverse changes in latent space, improving generalization and performance across various benchmarks.
Why It Matters
This research addresses the limitations of existing unsupervised change detection methods that rely on predefined assumptions, which can hinder their effectiveness in real-world applications. By leveraging latent space perturbations, MaSoN enhances adaptability and accuracy in detecting changes, making it a significant advancement in remote sensing technology.
Key Takeaways
- MaSoN synthesizes diverse changes in latent feature space, improving generalization.
- The framework outperforms existing methods by 14.1 percentage points in average F1 score.
- It is adaptable to new modalities, such as Synthetic Aperture Radar (SAR).
- The approach eliminates reliance on handcrafted rules or external datasets.
- MaSoN demonstrates state-of-the-art performance across five benchmarks.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.19881 (cs) [Submitted on 23 Feb 2026] Title:Make Some Noise: Unsupervised Remote Sensing Change Detection Using Latent Space Perturbations Authors:Blaž Rolih, Matic Fučka, Filip Wolf, Luka Čehovin Zajc View a PDF of the paper titled Make Some Noise: Unsupervised Remote Sensing Change Detection Using Latent Space Perturbations, by Bla\v{z} Rolih and 3 other authors View PDF HTML (experimental) Abstract: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 extend...
