Blind Image Quality Assessment via Multiperspective Consistency

Algorithm 1

Multiperspective Consistency-Based Model Training.

	Input:, learning rate and , Epochs,
	Output:
	Initialize parameters of and : initialize with pretrained parameters on ImageNet, initialize randomly; ;
	whiledo
	while Fetch minibatch from do
	Compute parameters using equation (3);
	Compute for images in minibatch using equation (6);
	Compute using equations (10) and (11), respectively;
	Compute the gradient and update and using Adam optimizer as equations:
	end
	;
	end
	return ;