"NA-CD Dataset (Non-Aligned Change Detection Dataset)"

"NA-CD Dataset (Non-Aligned Change Detection Dataset)Due to the lack of publicly available datasets for non-aligned change detection, we construct the NA-CD dataset. In the data generation process, the misalignment of the input data\u2014represented by the affine transformation matrix $A_{23}$\u2014is artificially introduced. To demonstrate the generality of our approach, the affine transformations span a wide range of translations, rotations, and scalings, effectively simulating the geometric misalignment caused by platform motion and multi-source acquisition in real-world scenarios.Different data generation strategies are adopted according to the source datasets:DOTA 1.5 Dataset: Random regions are selected and corrupted with noise or blur to create synthetic change regions, allowing evaluation of the model's ability to recognize image degradation types. Subsequently, HSV and affine perturbations are applied to generate the pre-change imagery.Google Dataset: Following the provided change region coordinates, gray-pixel filling is applied to the change areas in the pre-image while preserving the original HSV values. Affine perturbations are then applied to synthesize the pre-change imagery.LEVIR-CD and WHU-BCD Datasets: No masking operation is required; object-based slicing and affine transformations alone are sufficient to produce the converted datasets.This dataset provides a high-quality benchmark for evaluating change detection algorithms under non-aligned conditions, enabling robust assessment of model performance under complex geometric distortions."