Aviv Moshe Albeg
Reputation: 131
How to decide what layers to use for content and style loss in style transfer?
When you choose a pretrained network for style transfer to determine style and content loss you need to decide which layers best represent the perceived style and content and use them to make said losses. How does one go about choosing such layers?
Upvotes: 0
Views: 223
Answers (1)
jiawei hu
Reputation: 41
As far as I know, the way a fully convolutional NN works is that the shallow CNN layers are capturing more simple features such as lines and squares, and the deeper ones are combining them to capture more complex ones(faces, shoes, etc). So for feature caption, you may want to use a deeper layer. When it comes to style, as you might know, it usually uses a combination of different layers.
My two cents.
Upvotes: 1
Related Questions
- How to use transfer learning with CNNs using images with multiple channels - hand engineered
- How transfer learning on EfficientNets work for grayscale images?
- Transfer learning with Keras. What should the final layer activation be?
- Minimize the small loss to zero loss in transfer learning
- How does backpropagation changes pixels (input data) in neural style transfer?
- What is best practice for which CNN fully-connected layers to keep when doing transfer-learning?
- Is it possible to perform style transfer on my own images?
- Can someone please explain the content loss function?
- Use of L2-normalization in style transfer - no weights involved?
- Keras component selection in loss calculation