How to input fractional score labels for each input image while training a CNN in pytorch for regression problem

During a convolution neural network training, each input image is assigned to a score (floating point). How to feed that fraction value corresponding to each image as a label while training a cnn? Then I want to use the MSELoss.

For normal classification one have to create one folder for each classes then store the respective images to that class folder but for a continuous fraction score label how to do that and what torch API to use? Any suggestions on this will be highly helpful.