How to change feature extractor to Resnet

[dennissetiawan]

Hey, thankyou for the provided codebase

Could you please provide steps to change the feature extractor with Resnet in the current repo?

Thankyou

[dennissetiawan]

@ussaema
I tried to change the feature extractor to resnet, but it seems after 10 epoch the generator could not learn well in generating the caption. Why and how can we achieve a good result with resnet with the provided codebase or there are something else to be changed beside the 'resnet' feature extractor parameter in generator.py?

[chun178]

Could you please provide steps to change the feature extractor with Resnet in the current repo?

Hello, I also want to modify to Resnet network, but I don't know where to start and how to operate it. Can you provide some suggestions? thanks

[dennissetiawan]

you can change the features_extractor='vgg' to features_extractor='resnet' in train.py

[chun178]

you can change the features_extractor='vgg' to features_extractor='resnet' in train.py
Thank you for your answer. I just found out that I need to modify the feature extractor in train.py, but during training, I found that the loss of the generator was much greater than using vgg. Do I think using Resnet should be lower, or is 10 epochs too little for Resnet training?