image stacking

Should participants take care of the image stacking part in their algorithm (.tif) as it has been done in the example algorithm, or they can easily load images (.jpg) in their code?

Both. They can build any container themselves, with the required input/output. However, to make things easier they can/should use the example helper.py and inference.py since these generate .jpgs from the stack itself.

What do you mean exactly by stacking? Multiple RGB images in on tiff file? O simply a single RGB image in a tiff file? In the case of just a single RGB image, the provided example code simply saves each channel as a separate JPG file, which is completely unnecessary.

Why simply not provide all test set images in the same JPG format as the training dataset?