Revisiting Consistency Regularization for Semi-supervised Change Detection in Remote Sensing Images
[IEEE TGRS 2025] Be the Change You Want to See: Revisiting Remote Sensing Change Detection Practices
[IEEE TGRS 2026] Make Some Noise: Unsupervised Remote Sensing Change Detection Using Latent Space Perturbations
Official implementation of "Risk-Controlled Urban Change Detection: Conformal Prediction Wrappers for Provable Reliability in High-Resolution Satellite Imagery" using PyTorch.
U-Net architecture for detecting building construction changes in satellite imagery
Lightweight Siamese Network for Remote Sensing Building Change Detection | 轻量化孪生网络遥感建筑物变化检测（LEVIR-CD 数据集）
Where did buildings appear between two satellite images? Attention U-Net (PyTorch) vs Random Forest on LEVIR-CD image pairs: test F1 0.59 vs 0.32. The trained model is then applied to Sentinel-2 RGB images of the 24 OSCD cities. Kaggle notebook, built with Kshitij Saxena.
A human-in-the-loop evaluation platform for reviewing Qwen2-VL outputs on bi-temporal satellite imagery. Built for the LEVIR-CD change detection dataset.
LiteCDNet 公开仓库 | Public repo for training, evaluation, and ablation experiments
Lightweight CNN change detector with a distribution-free false-discovery guarantee. Model-agnostic conformal layer applies to any change detector.
Can frozen DINOv3 satellite embeddings find new buildings? Temporal transformer on SpaceNet-7 and LEVIR-CD, free Colab T4. Test F1 0.91 on LEVIR-CD, where a 0.56M-parameter difference head matched a cross-attention head 1.5x its size; 0.18 on SpaceNet-7's much harder monthly task, against 0.02 for raw embedding distance.
To associate your repository with the levir-cd topic, visit your repo's landing page and select "manage topics."