pengxiao-wang/style2paints_v3

Reimplementation of Style2Paints V3

style2paints_v3's Issues

model in download_patterns.sh is 404 now

imagenet_224_vgg_16

Where is the mod.h5

Where is pre-trained UNet model?

Hi, I would like to test your model.
but I can't find pre-trained UNet model.
It seems that Unet weights is used in your code in Pytorch-Style2paints/test.py (maybe 25_Unet_checkpoint.pth.tar). Could you share the model?

您好，我是初学者，请问训练数据怎么弄的

'train_path' : 'train data path'#'./your train data path/train',
'val_path' : 'val data path'#'./your val data path/val',
'sketch_path' : 'sketch path'#"./your sketch data path/sketch",
'draft_path' : 'STL path'#"./your STL data path/STL",

我知道训练数据来自Danbooru，但是不知道 train,val,sketch,STL里分别是什么

关于边缘检测时用到的模型文件的需求

你好，我想用test.py测试一下我根据你这个项目训练出来模型的效果，不过我缺少下面代码中所缺少的mod.h5文件，你能够发一下或者给一下相关下载链接吗
load_model('/data2/wangpengxiao/GANs/style2paints3/V4/mod.h5')

sketchkeras's sketches should be purified

I notice that you didn't purify the sketchkeras's sketches for training.Though,raw sketchkeras's sketches contain too match information for unet to reconstruct image ,that's why your effect are pleasing?

图片数量

我用了danbooru两万五千张图片做训练，效果非常差，想知道到达图示的效果用多少张图片？
另外直接用百度网盘中的11_Unet_checkpoint.pth.tar无法运行,需要将unet_model.py里面forward函数中的x_inception = self.avg_pool(x_inception)解注释，想问下还有什么需要注意的点吗？

在测试的时候为什么要将df resize成299大小的？

在运行Unet(input_var,df_var)过程当中，x9 和 x_inception尺寸大小不同，x9：（1，1024，16，16）；x_inception （1，1024，4，4）。我的input_var的大小是（1，4，256，256）

Where should I download the dataset?

Thanks for your contribution. The original tensorflow source became invalid, and I searched the project of style2paints in Github but didn't find the way to download it. Could you tell me the resource of dataset I can download?