Closed premtibadiya closed 3 years ago
I dont have small GPU But I am trying transfer learning and where is the code which loads the weights trained from IAM dataset
even on CPU is is trained in less than a day.
If you want to transfer the weights, you have to read the TF documentation how to do this. I simply use the function tf.train.latest_checkpoint
to load the latest checkpoint file, that means, TF cares about all the details.
retrain from scratch. It is really not worth the effort doing any work in this direction for a model that trains in 3h even on a small GPU.