尊敬的作者大大你好，关于训练的问题有一些不明白

[tachibanayoshino]

尊敬的作者大大你好，关于训练的问题有一些不明白

[TachibanaYoshino]

Hi,
If you want to directly use the pre-trained model for style transfer, please run test.py for picture transfer and video2anime.py for for video transfer directly.
If you want to train your own model, you can follow the steps. Do edge_smooth is a necessary step, which is used during training. The channel(BGR) color difference is not necessary, it is only the preprocessing of the data, I will delete this step.
Regarding the dark and explosive scenes, it may be related to the problem you mentioned.
If you have enough GPU memory space, you can use larger resolution images for training, that is, you can customize the shape of the input image according to your dataset.

[thefreeman007]

Thank you for your answer. Do edge_ Is the smooth step prediction also necessary？
I'm very sorry I asked questions in Chinese. I thought you were my Chinese compatriots. Thank you for your answer. If I train a better model, I will share it with you at the first time

[TachibanaYoshino]

In the inference phase, Do edge_smooth is not needed. Do edge_smooth is the preprocessing of the animation dataset, which is used for edge promotion loss during training.

[thefreeman007]

Thank you very much. You are a kind and handsome man.

[1054171582]

anavanab99 同学我可以加你的联系方式吗，因为我也再接触这个项目我有一些问题不太明白想和您请教 微信13051093600 希望看到加我一下谢谢

[aluds123]

@TachibanaYoshino , Hi, I have some questions want to ask you:
"since the real photos in the training set are all landscape photos, if you want to stylize the photos with people as the main body, you may as well add at least 3000 photos of people in the training set and retrain to obtain a new model."
=>

1. Do the photos of people in the training set also need to do "edge_smooth"?
2. The 3000 photos of people should be put into "AnimeGAN-master\dataset\train_photo" folder, right?

Thank you.