Currently, only 16-bit images in png format are supported. As internal computation of core library is performed in 32-bit floating point, using 16-bit png images for training and validation data allows training and evaluation in 16-bit resolution. Even when training is performed using 16-bit images, the evaluation images shown on the EVALUATION tab are 8 bit.
Data with the same variable name (e.g., x) must be of the same size. Different variables (e.g., x and y) can use different sizes.