search

lidq92 / LinearityIQA

[official] Norm-in-Norm Loss with Faster Convergence and Better Performance for Image Quality Assessment (ACM MM 2020)
https://lidq92.github.io/LinearityIQA/
97 stars 14 forks source link

The result of cosine_similarity in IQALoss #12

Closed shuishiwojiade closed 3 years ago

shuishiwojiade commented 3 years ago

Hello, Thank you for sharing the great work. But I have a confusion in 109th line of code in IQA Outputs and Targets are tensors of B * 1, so the value of the element in the rho is 1 or -1

lidq92 commented 3 years ago

@shuishiwojiade Oh, it is a wrong implementation. Thank you for pointing it out! The correct one should be rho = torch.cosine_similarity(y_pred.t(), y.t()). This may affect the result of 𝑙 + 0.1𝑙′ in Table 2. I have corrected this code (See https://github.com/lidq92/LinearityIQA/commit/7936bb811ae9409f1858e4d8f558d8153091edcf) and have updated the result of the combination of 𝑙 and 𝑙′ in this repo (See https://github.com/lidq92/LinearityIQA/commit/b2f10477aca9e6cf307ad86297759d0b7c308470). Thank you again.