Could you explain about the dataset?Posted by: sabarinathan @ May 20, 2020, 8:55 a.m.
Is there something specific that was unclear? The dataset is presented on the "Participate"/"Get Data" page, and the data are the same as that in track 2 but downsampled by 2.
Generally what you need to know is that you get a pair of images (Input, Guide) and you need to produce/generate/synthesize (Target). (Target) should have the content of (Input) and the illumination settings of (Guide).
For training you can form any triplet (Input, Guide, Target) from the training dataset since you are given the images under all illumination settings. If you select as train Input Image000_2500_N.png, you can pick a train Guide image for example Image111_4500_W.png, then the corresponding target would be Image000_4500_W.png (notice it is the content of Image000 but with the illumination settings 4500_W instead of 2500_N)
For validation/test we provide (Input, Guide) pairs and you produce (Target).
Thank you for the clear explanation.Posted by: sabarinathan @ May 20, 2020, 11:30 a.m.