CCAH: A CLIP-Based Cycle Alignment Hashing Method for Unsupervised Vision-Text Retrieval

Algorithm 1

CLIP-based cycle alignment hashing for unsupervised vision-text retrieval.

	Require: Image set ; text set ;
	Batch size set , hash code length , Max epoch .
	Ensure: Deep Feature extract functions , and ;
	encoder function set , and ;
	Hash coding functions , and .
(1)	Initialize the pretrained extractor parameters: .
(2)	Whiledo
(3)	;
(4)	Extract the depth characteristics of each mode: , ;
(5)	Encode the features to get the hidden states, by ;
(6)	Using the hidden states to generate truth matrix and hash codes;
(7)	Decode the hidden states to generate heterogeneous features and
(8)	Calculate the objective function;
(9)	Back propagate the gradient with the chain rule;
(10)	Update the whole parameters;
(11)	end while