TypeError: '>' not supported between instances of 'str' and 'int' about homointerpgan HOT 6 OPEN

Hi @PeterouZh, I think the reason might be that f3 parse "-1" as string instead of int. Can you try to replace that line with f3 = self.frame.replace(-1, 0) or f3 = self.frame.replace("-1", 0)? Please let me know whether it works or not.

from homointerpgan.

Hi @yingcong , I'm facing the same problem, and I tried your method to solve it. But after change that line, it keeps run time error at the start of training(showing "stack overflow")
Then follows lots of lines of "***********/data/attributeDataset.py", line 244 in __getitem__"
Here's some first lines of the error message.

loading default VGG
loading default VGG
loading default VGG
* Total Images: 162770
* Total Images: 19962
save_dir: checkpoints/CelebA_my
n_discrim: 5
command: train
gpu: 0
batch_size: 128
data_dir: ../celebra-dataset
attr: Mouth_Slightly_Open@Smiling,Male@No_Beard@Mustache@Goatee@Sideburns,Black_Hair@Blond_Hair@Brown_Hair@Gray_Hair,Bald@Receding_Hairline@Bangs,Young
dec_type: v1
continue_train: False
epoch: 10000

  0%|                                                                                                                                                                        | 0/1271 [00:00<?, ?it/s]Fatal Python error: Cannot recover from stack overflow.

Current thread 0x00007f8675a0b740 (most recent call first):
  File "/....../HomoInterpGAN/util/util.py", line 144 in readRGB
  File "/....../HomoInterpGAN/data/attributeDataset.py", line 244 in __getitem__
  File "/....../HomoInterpGAN/data/attributeDataset.py", line 266 in __getitem__
  File "/....../HomoInterpGAN/data/attributeDataset.py", line 266 in __getitem__
  File "/....../HomoInterpGAN/data/attributeDataset.py", line 266 in __getitem__
  File "/....../HomoInterpGAN/data/attributeDataset.py", line 266 in __getitem__
......

I'm using pytorch 1.0.1 with cuda 9.1, other modules are installed as requirement.txt
I'm looking forward to your help, thanks!

from homointerpgan.

Hi @oscardhc It seems that images are not successfully loaded with getitem. Have you checked if the celeba images (jpg format) are in the ../celebra-dataset file? If yes, could you remove the "try -- except" block in the getitem method? This could give more information about why images cannot be loaded.

from homointerpgan.

@yingcong Thank you for the advice! It turns out that I've put all images in a subdirectory in the dataset path(as the default result of unzipping). After moving the images, your solutions above work and it starts to train.

from homointerpgan.

I fix this issue by replacing "f3 = self.frame.iloc[:, 1:] > 0" url with "f3 = self.frame.iloc[:, 3:-1] > 0".

from homointerpgan.

Hi @PeterouZh I've replaced f3 = self.frame.iloc[:, 1:] > 0 with f3 = self.frame.replace(-1, 0) in the later commit, which seems more general for different pandas versions.

from homointerpgan.