Hi!
Interesting work! However, I have some confusion about Acc_g. In the paper, Acc_g refers to the accuracy of synonyms, but in the code, when calculating the similarity between images and objects, it goes through softmax. Due to different object lists, the results might be different. However, there's a default object list with 10 classes, which doesn't seem to include the listed synonyms. I would like to know what the specific method is when calculating Acc_g?
Thanks!
Shilin-LU
commented
4 months ago
Hi! Interesting work! However, I have some confusion about Acc_g. In the paper, Acc_g refers to the accuracy of synonyms, but in the code, when calculating the similarity between images and objects, it goes through softmax. Due to different object lists, the results might be different. However, there's a default object list with 10 classes, which doesn't seem to include the listed synonyms. I would like to know what the specific method is when calculating Acc_g? Thanks!