I don't feel safe about what the task is so I want to make the task clear.
Overview says "At runtime, it is assumed as input a low spatial resolution spectral image and its aligned colour image in high spatial resolution."
So I think that we have to train a model F that predicts high-resolution spectral image(y1) and high-resolution appealing-to-human pseudo color image(y2) from inputs low-resolution spectral image(x1) and high-resolution aligned color image(x2) , y1, y2 = F(x1, x2), x1, x2, y1, y2 are from the viewpoint of spectral camera.
However readme.pdf in dataset says test_hr.zip is available when the challenge has ended. So I am confused. Please make things clear.
Thank you.
Posted by: nattochaduke @ July 18, 2018, 2:42 a.m.Hi,
Please note that the RGB images at resolution say lr2, has 4 times the resolution of hyperspectral images at the same level (lr2). This means that the width and height of RGB images are twice higher than hyperspectral images.
So, test_lr will contain lr2 and lr3 level hyperspectral and RGB images. The RGB images are 4 times the resolution of the hyperspectral images.
For the physical cameras, the RGB sensor and the IMEC sensor are with equal raw spatial resolution. The image sensor for the RGB camera with width = W and lengths = L for the raw image has a Bayer pattern of RGGB (it is not debayered yet); so the debayered RGB image should result in resolution of (W/2, L/2). The IMEC sensor with dimensions (W, L) will lead to hyperspectral images of resolution (W/4, L/4) due to the 4x4 filter array.
Posted by: mehrdad.shoeiby @ July 19, 2018, 12:07 a.m.I understood. Thank you.
Posted by: nattochaduke @ July 20, 2018, 6:03 a.m.