SDGSAT-1 Misalignment dataset for Object Detection

Annotated 1,000 misalignment from the SDGSAT-1 glimmer imagery, divided into train, valid, and test sets with a ratio of 7:2:1 for the object detection task.
This dataset contains only one type of object: misalignment. We used a 32×32 window to crop the raw SDGSAT-1 Level-1 glimmer imagery and converted the TIFF format to JPEG format. At each window, a column number was randomly selected, and the corresponding pixels to the right of this column were shifted vertically either upward or downward by 2 to 8 pixels. The annotations were done in COCO format using LabelImg, with each TXT label file corresponding one-to-one with the JPEG image files.