You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, the test images are randomly selected from the rest of dataset, so you can easily generate the 'test.txt' by yourself just referencing to 'train.txt'.
Thank you for your reply! and I used the mini-batch=128 in Flickr datasets to construct the similarity matrix,but I found the percentage of 1 is about 0.7% and the percentage of 0 is about 44% in similarity matrix, and then paper uses this sim matrix to construct negative log-likelyhood function,whether it will cause the imbalance of positive and negative samples?
Can you provide the test.txt ?
The text was updated successfully, but these errors were encountered: