-
git clone源码
git clone https://github.com/hDluffy/stylegan2.git
同时下载源github中提供的预训练模型stylegan2-ffhq-config-f.pkl,放置networks目录下
-
安装训练环境
conda配置
conda create --name stylegan2-tf python=3.6 source activate stylegan2-tf ##注意cuda版本与系统安装版本无关 conda install tensorflow-gpu=1.14 pip install scipy==1.3.3 pip install requests==2.22.0 pip install Pillow==6.2.1 pip install typer
docker环境
docker build .
-
数据处理
1024x1024对齐后的raw数据,通过dataset_tool.py转为tfrecords数据集
#转为tfrecords数据集 ./make_dataset.sh #转为tfrecords数据集的具体指令 python dataset_tool.py create_from_images ~/datasets/my-custom-dataset ~/my-custom-images #可以查看数据 python dataset_tool.py display ~/datasets/my-custom-dataset
-
模型训练
./train.sh #CUDA_VISIBLE_DEVICES=0,1,2,3 python run_training.py --num-gpus=4 --data-dir=~/datasets --config=config-f --dataset=my-custom-dataset --mirror-augment=true
注:
可能的错误undefined symbol: _ZN10tensorflow12OpDefBuilder6OutputESs
解决方案:https://blog.csdn.net/zaf0516/article/details/103618601
可能的错误#error "C++ versions less than C++11 are not supported
解决方案:https://blog.csdn.net/qq1483661204/article/details/1054424261.通过设置./traing/training_loop.py中的resume_pkl路径,可以修改预训练模型
2.建议注释掉评估,可以加快训练【metrics】 -
测试及批量处理
#task='GenShow',显示不同层融合的效果对比;task='GenBatch',进行特定层融合效果的批量生成; ./gen_images.sh #处理完的数据,可以通过指令copy到本地处理 scp username@ip:/path/data /mnt/g/DataSets
forked from NVlabs/stylegan2
-
Notifications
You must be signed in to change notification settings - Fork 0
StyleGAN2 - Official TensorFlow Implementation
License
hDluffy/stylegan2
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
StyleGAN2 - Official TensorFlow Implementation
Resources
License
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published
Languages
- Python 93.9%
- Cuda 5.8%
- Other 0.3%