Skip to content

Latest commit

 

History

History
15 lines (13 loc) · 1.25 KB

README.md

File metadata and controls

15 lines (13 loc) · 1.25 KB
Raw

OpenFaaS Style Transfer (Avatar-Net: Multi-scale Zero-shot Style Transfer by Feature Decoration. CVPR-2018)

Check result

sudo docker run -it --net="host"  -e STYLE_PATH='styles/candy.jpg' -e INTER_WEIGHT=0.5 dfkozlov/openfaas-style-transfer
curl --request POST --url http://localhost:5000 --data-binary @"image.jpg" --output style_transfer_result.jpg

Build docker image

  • The trained model of Avatar-Net can be downloaded through the Google Drive.
  • Pre-trained AvatarNet models, place them in the ./AvatarNet folder and create a file named "checkpoint" in the same directory (./AvatarNet) with the three .index, .data, .meta files. The file "checkpoint" (no extension) has the line model_checkpoint_path: "model.ckpt-120000".
  • The encoding layers of Avatar-Net are also borrowed from pretrained VGG model. Place the vgg_19 folder (with only vgg_19.ckpt inside) in the avatar-net/models folder.
  • In AvatarNet_config.yml, change the checkpoint_path on line 10 to point to the vgg_19.ckpt file. checkpoint_path: '../models/VGG/vgg_19.ckpt'