Questions about the fusion model

[lvshuai789]

Hi, kualee: I am a doctoral candidate in Beijing University of Posts and Telecommunications. I am also currently working on cross-modal graphic retrieval directions and it is my great honor to read the paper SCAN. I have some questions of my own to ask you，I wonder if it is convenient for you.First, for the experimental results，as shown in the picture below:

My understanding is: The I2T-trained model can deduce i2t and t2i indexes on f30k respectively Then the T2I-trained model can also deduce the i2t and t2i indexes separately on f30k Where did i2t + t2i come from?