pfnet-research / chainer-pix2pix Goto Github PK

I run your code with slightly different network parameters:

enc = Encoder(in_ch=1)
dec = Decoder(out_ch=37)
dis = Discriminator(in_ch=1, out_ch=37)

After many training iterations I found that only the first channel out of the 37 channels of "out_ch" converges. When checking the "gen_all" output at your out_image module the first channel converged fine but all other channels have values between -1 and -0.8.

Would it be anything else to have in mind when changing encoder's, decoder's and discriminator's channels?

Would it be possible the case that pix2pix cannot learn to produce more output channel than inputs channels?

**UPDATED
Few more notes:
-When tested all other parameters such as network configuration and image dimensions were kept the same
-Dataset had 35000 samples
-The first layer converges impresively well after 4.5 epochs

Any help/advice will be really appreciated.

Thanks,
Alex

At line 84 in updater.py, do you actually need to unchain x_in given that x_in does not have any creator (basically x_in is created by the user so that it has no reference)?

Hi,
Thanks for publishing valuable program!
However, I got stuck in the exception problem.

I got the error below on running program.
Does this mean this program no longer work with chainer 2.0.0?
Any reply would be helpful.

python train_facade.py --dataset "./facade/base" --out "./out" --snapshot_interval 10000
GPU: -1
# Minibatch-size: 1
# epoch: 200

load dataset start
    from: ./facade/base
    range: [1, 300)
load dataset done
load dataset start
    from: ./facade/base
    range: [300, 379)
load dataset done
Traceback (most recent call last):
  File "train_facade.py", line 119, in <module>
    main()
  File "train_facade.py", line 116, in main
    trainer.run()
  File "/Users/kkbnart/.pyenv/versions/dcgan/lib/python3.4/site-packages/chainer/training/trainer.py", line 296, in run
    update()
  File "/Users/kkbnart/.pyenv/versions/dcgan/lib/python3.4/site-packages/chainer/training/updater.py", line 177, in update
    self.update_core()
  File "/Users/kkbnart/Documents/programming/python/machineLearning/pix2pix/chainer-pix2pix/updater.py", line 73, in update_core
    z = enc(x_in, test=False)
  File "/Users/kkbnart/Documents/programming/python/machineLearning/pix2pix/chainer-pix2pix/net.py", line 58, in __call__
    hs.append(self['c%d'%i](hs[i-1], test=test))
  File "/Users/kkbnart/Documents/programming/python/machineLearning/pix2pix/chainer-pix2pix/net.py", line 33, in __call__
    h = self.batchnorm(h, test=test)
  File "/Users/kkbnart/.pyenv/versions/dcgan/lib/python3.4/site-packages/chainer/links/normalization/batch_normalization.py", line 117, in __call__
    kwargs, test='test argument is not supported anymore. '
  File "/Users/kkbnart/.pyenv/versions/dcgan/lib/python3.4/site-packages/chainer/utils/argument.py", line 4, in check_unexpected_kwargs
    raise ValueError(message)
ValueError: test argument is not supported anymore. Use chainer.using_config

Hi,
Here is a problem when I tried to run train_facade.py (I have downloaded the dataset). Can you help me with this problem? Thanks :)

“Traceback (most recent call last):
File "/home/wh/wh/chainer-pix2pix-master/train_facade.py", line 119, in
main()
File "/home/wh/wh/chainer-pix2pix-master/train_facade.py", line 70, in main
train_d = FacadeDataset(args.dataset, data_range=(1,300))
File "/home/wh/wh/chainer-pix2pix-master/facade_dataset.py", line 39, in init
img = np.asarray(img).astype("float")
File "/usr/lib/python2.7/dist-packages/PIL/Image.py", line 528, in getattr
raise AttributeError(name)
AttributeError: float”