Closed ofirkris closed 4 years ago
Hi, do you mean the content, warp, and regularization weights? I personally like using content_weight=8, warp_weight=0.5, and reg_weight=50 as I put in job_example.sh
!
The style transfer part optimizes online, it does not need any trained weights. (that's why it takes about 6 minutes to generate an image) The feature extraction part requires imagenet pretrained VGGnet weights, which will be automatically downloaded by Pytorch when running for the first time.
Hello, Amazing work! Can you share some trained weights for evaluation and testing purpose?