komiya-m / mirrorgan Goto Github PK

When I run the pretrain_STREAM.py, the file "model/cnn_rnn_encoder02_.h5" is not found. How should I get the model file? Thank you.

How can we use gpu for training this model as it is training very slowly on CPU.

Hello, when I run the code in Pycharm, I get the following error:

Traceback (most recent call last):
  File "pretrain_STREAM.py", line 89, in <module>
    main()
  File "pretrain_STREAM.py", line 47, in main
    assert dataset
AssertionError

I believe it is due to the function load_filenames (posted below). It is from dataset.py. Starting on line 235. Whenever the program reaches this statement in the function:

if os.path.isfile(filepath):

It always evaluates to false. When this happens I get the above error. In both cases when the function is called, it seems to be looking for a "filenames.pickle" file that doesn't exist. The files are supposed to be in:

data/birds/train
data/birds/test

My question is, are these files supposed to exist before running the code, or does the code create them during runtime?

I noticed that the code created "captions.pickle". So I wasn't sure if they should be doing the same with the other two. Thank you.

   def load_filenames(self, data_dir, split):
        filepath = '%s/%s/filenames.pickle' % (data_dir, split)
        if os.path.isfile(filepath):
            with open(filepath, 'rb') as f:
                filenames = pickle.load(f)
            print('Load filenames from: %s (%d)' % (filepath, len(filenames)))
        else:
            filenames = []
        return filenames

Help me please,,,,,,How to adjust the parameters.
I only get 4.16 by using this code

100%|██████████| 442/442 [13:11<00:00, 1.83s/it]
D_loss: 0.21999 D_wrong_loss: 1.60813 D_acc: 0.96270 D_wrong_acc: 0.00251
G_loss: 1.08059 G_discriminator_loss: 0.22026 G_encoder_loss: 8.60321
0%| | 0/442 [00:00<?, ?it/s]----------------EPOCH: 11 START----------------
100%|██████████| 442/442 [13:02<00:00, 1.89s/it]
D_loss: 0.21901 D_wrong_loss: 1.60871 D_acc: 0.96179 D_wrong_acc: 0.00501
G_loss: 1.04197 G_discriminator_loss: 0.18165 G_encoder_loss: 8.60318

This is my reslut, seems like the G_ENCODER_LOSS didn' change.. is there any has the same problem or what should I do for this???

OSError: Unable to open file (unable to open file: name = 'model/cnn_rnn_encoder02_2.h5', errno = 2, error message = 'No such file or directory', flags = 0, o_flags = 0)

i do not get this cnn_rnn_encoder02_.h5 file in the master code ,so how can I get this file

in lines 124~128 train.py, why fake label are set as the ground truth for the first output, considering the real_image is the input?
if batch % wrong_step == 0:
histDw = D_model.train_on_batch(
[real_image[:-1], captions_ar_prezeropad[1:]],
[fake_label[:-1], fake_label[:-1]],
)