languandong

some question about the position of 'optimizer.zero_grad()'

4

I think the correct way the code the training is that optimizer.zero_grad() # Forward pass outputs = model(images) loss = criterion(outputs, labels) # Backward and optimize loss.backward() optimizer.step() not that...

for example.I just want to get the coordinates of No.3 and No.4 point?What should I do?

checkpoints didn't perform as what paper shown

7

Excellent work! As for VQA-RAD dataset of the VQA task, I download the checkpoint you provided and run the test script. The effect was shown below.  But paper's 

I don't know how to complete the download?

5

I run the script "python scripts/fetch.py". But it downloaded as 9.895%.  So I immediately re-executed the script.  Should I take -d at the end of this instruction?