Hi, I want to train a networks for predicting the quality of an image (Image Quality Assessment). Hence, the output will be a single value. I've carefully checked the issues and already found useful information about how to implement the necessary customized loss layer such as L2-loss, L1-loss etc. However, it still not clear how should I change the error function in cnn_train.m.( Obviously, the error function should be neither the case of 'multi-class' nor 'binary-class'). Intuitively, the error should be the difference between the predicted quality score of the input image and the label (real quality score of this image). Any help would be deeply appreciated !
Hi, I want to train a networks for predicting the quality of an image (Image Quality Assessment). Hence, the output will be a single value. I've carefully checked the issues and already found useful information about how to implement the necessary customized loss layer such as L2-loss, L1-loss etc. However, it still not clear how should I change the error function in cnn_train.m.( Obviously, the error function should be neither the case of 'multi-class' nor 'binary-class'). Intuitively, the error should be the difference between the predicted quality score of the input image and the label (real quality score of this image). Any help would be deeply appreciated !