StereoMSI

1. Multi-spectral super-resolution dataset (extension to PIRM2018 multi-spectral super-resolution challenge dataset).\nThe dataset includes\n- 272 training stereo registered multi-spectral RGB image pairs including the original RGB mosaic images.\n- 970 raw multi-spectral images. These need to be downsampled (x2 or x3 etc.) to create low-resolution images for training a CNN for the task of super-resolution.\n- 50 test images.\n2. Dataset for simultaneous colour-prediction and super-resolution.\nThis dataset includes:\n- 250 registered stereo multi-spectral/RGB image pairs for training\n- 25 registered stereo multi-spectral/RGB image pairs for validation\n- 21 registered stereo multi-spectral/RGB image pairs for testing.\nLineage: 1. Images in the multi-spectral super-resolution dataset are in .tiff format.\n2. Images in the colour-prediction dataset are in .tiff (RGB) and .mat (multi-spectral) formats. Converting .mat to .npy can accelerate training. See https://arxiv.org/pdf/1909.02221.pdf for more information on multi-spectral data format.