Beginner question about frozen layers and transfer learning

[emipasat]

When training the RPN Model for example, wouldn't be more suitable to reuse the weights for 'imagenet' instead of retrain all model from scratch? Did you noticed better results when training all model instead of only last layers? Thank you

[FurkanOM]

Hi @emipasat,

1. The model is already using imagenet weights. You need to read about default params of keras applications, for VGG16.
2. The model is already working to train not only the last layers but all the layers.

[emipasat]

Sorry for not being clear. My question was if is worth to freeze the backbone layers and to train only the ones added by you, the last ones. Would it work that, too? Thank you

[FurkanOM]

You could give it a try. However, in my experience, fine-tuning generally gives better results than transfer learning.