python == 3.6
tensorflow == 1.10.0
numpy == 1.14.5
Frist, download all dependent files.
Please download slim vgg_16 check point from http://download.tensorflow.org/models/vgg_16_2016_08_28.tar.gz
Please download coco dataset from http://images.cocodataset.org/zips/val2017.zip
wget http://download.tensorflow.org/models/vgg_16_2016_08_28.tar.gz
wget http://images.cocodataset.org/zips/val2017.zip
Second, fit your transfer model.
python train.py --vgg_path "your vgg pre-trained path" --model_path "your transfer model path"
--style_image_path "choose one style image" --train_dataset "coco dataset path" --epoch 20
--batch_size 4 --learning_rate 0.001 --content_loss_weight 1.0 --style_loss_weight 100.0 --summary_path "summary path"
Third, you can use tensorboard to visual training process. you should access http://127.0.0.1:8888
tensorboard --logdir "your summary path" --port 8888
Last, you can use trained model to inference.
python inference.py --test_image "test image" --model_path "your transfer model path" --saved_path "transfromed image path"
NOTE:
1.I provide my trained model in this link: https://pan.baidu.com/s/1vgNb1sdfxVYXwLAtrv0ndw
2.You should fine tune the style_loss_weight, if you want to get a different style weights.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent