Researchers analyzed the subimages to detect subtle variations in lighting across the panorama.

The machine learning model extracts each subimage to analyze local features.