Code errors

[Johnson-yue]

Hi, your data_loader.py have some bugs ,such as :

1. in pretrain() function, data_loader should be return 7 element, but ImageFolder() only return 5 elements. I think , in pretrain() function, the data_loader should be EncoderFolder() not ImageFolder(), please check your code.

2. In pretrain() function. self.E(src_img) should be return x_sout, x_cout, cls_style, cls_char, but your Encode() model return feature and None , so the bug at line 223 in solver.py. I think self.E should be Class ClassifierList not be Class EncoderList

[yongqyu]

Thank you for your attention.

I modified the 145th line of manually data_loader.py for pre-train.

To reduce the memory usage, we have created a classifier class like Encoder. Classifier class should be used in pre-training, and Encoder class is used in main-training.
I apologize for the need for manual code modifications.

Thank you.

[Johnson-yue]

oh.....you run the code twice.
First is for pre-training, and manually modified with EncoderFolder() in data_loader.py
Second is main-training, and use ImageFolder() and EncoderList replace EncoderFolder() and ClassifierList, is right?

[yongqyu]

In pre-training,
dataloader : EncoderFolder()
Encoder : ClassifierList
In main-training,
dataloader : ImageFolder()
Encoder : EncoderList

it is right.

[Johnson-yue]

3. load E model error， I think it should be :
   E_path = os.path.join(self.model_save_dir, '{}-E.ckpt'.format(self.enc_iters))

4. I think your train dataset is very small，when I create my train dataset contain 20 typefaces and every typeface have 6763 characters，your progress() funtion is so slowly!!

5. In Classifier() model, you used nn.Conv2d with kernel_size=32, it cost large GPU memory, when I change batch_size is 2, It is also out of memory in my GTX1080ti, 11GB
   Why do you using the so large kernel ???