CrossmodalGroup/SSL-VQA

not understand code work

Opened this issue · 0 comments

in your code , train.py

if epoch < opt.pretrain_epoches:
    logits_pos, _= model(q, v, False)
else:
     logits_pos, logits_neg, _, _ = model(q, v, True)

why not use self-sup at the beginning?