Thank you for your outstanding research. I am a beginner in video quality assessment, and after carefully reading your paper and reviewing the code, I encountered a question. In your paper, you discuss using the loss function Lcr for cross-scale regularization, but when I checked the code, it seems that this loss function is not used in the training code. Specifically, the function def self_similarity_loss(f, f_hat, f_hat_detach=False): does not appear to implement it. Could you please clarify whether this loss function is used, and if so, where is it applied?
Thank you for your outstanding research. I am a beginner in video quality assessment, and after carefully reading your paper and reviewing the code, I encountered a question. In your paper, you discuss using the loss function Lcr for cross-scale regularization, but when I checked the code, it seems that this loss function is not used in the training code. Specifically, the function def self_similarity_loss(f, f_hat, f_hat_detach=False): does not appear to implement it. Could you please clarify whether this loss function is used, and if so, where is it applied?