Training speed (batch)

[wooni-github]

Thank you for your excellent research!

I’m trying to train the model using the code you provided (on the Cambridge Landmark dataset), but the training process is taking too long.

I considered increasing the batch_size in Cambridge.yaml under:

train:  # train configs
  batch_size: 1

However, as mentioned in this issue, increasing the batch size causes an error.

Is there another way to speed up the training process?

[thuanaislab]

Thank you for your interest!

Training time is currently one main drawback of this work.
If you don't increase the batch_size, then there isn't any option to speed it up.