pfnet-research / chainer-gan-lib Goto Github PK

View Code? Open in Web Editor NEW

408.0 34.0 84.0 71.22 MB

Chainer implementation of recent GAN variants

License: MIT License

Python 98.98% Shell 1.02%

deep-learning generative-adversarial-network dcgan wgan-gp

chainer-gan-lib's People

Contributors

Stargazers

Watchers

Forkers

hiroshiba oppa3109 codeaudit shihuaxing winnerineast mydp2017 mindis xunhuang1995 machinelearninghelen rhythm92 kingofoz knok iory canhunan1 3sunny takerum aixile h6ak linpingchuan soroushmehr shubhampachori12110095 rl-gan-vision-privacy-finance-projects izhangh picaosgeo winsonguo albertlzg kouui homli mr4msm liqunchen0606 iaroslav-ai huidingumd zhf459 levelfour ep3lb datagrid-okada csjunxu howardchina hmjianggatech claber zeitgeistqian yanshui177 heikir lllyasviel haiminzhang xtrigold maxliu0426 afcarl tettu1118 keiikegami mannykayy 127t6937 meliodas-legendary-super mahmud83 hengle yuyay uidilr zilinyan ryukimizo yuhonghong66 peachang gongxinyuu kylesargent alexzhwqc grayuranium hbdean96 tanmdl asker0lee dai7igarashi madonokouki tiankaiyuan johnchenresearch hellotem linleyzoe jjshisajia juhojokinen67 classicvalues forky-mcforkface shunsunsun iq-scm 5l1v3r1 zhaopufeng

chainer-gan-lib's Issues

ResNet based SN-DCGAN model for training ILRSVRC2012 dataset?

Will you release the code of ResNet based SN-DCGAN model for training ILRSVRC2012 dataset? I'm curious about the details of this network architecture. Thanks.

ValueError: volatile argument is not supported anymore. Use chainer.using_config

Traceback (most recent call last):
File "/media/scs4450/52d95cdd-64a0-4c37-9480-dc43778ffc5f/next/gan/chainer-gan-lib-master/train.py", line 191, in
main()
File "/media/scs4450/52d95cdd-64a0-4c37-9480-dc43778ffc5f/next/gan/chainer-gan-lib-master/train.py", line 187, in main
trainer.run()
File "/usr/local/lib/python3.4/dist-packages/chainer/training/trainer.py", line 313, in run
six.reraise(*sys.exc_info())
File "/usr/local/lib/python3.4/dist-packages/six.py", line 693, in reraise
raise value
File "/usr/local/lib/python3.4/dist-packages/chainer/training/trainer.py", line 302, in run
entry.extension(self)
File "/media/scs4450/52d95cdd-64a0-4c37-9480-dc43778ffc5f/next/gan/chainer-gan-lib-master/common/evaluation.py", line 104, in evaluation
mean, std = inception_score(model, ims)
File "/media/scs4450/52d95cdd-64a0-4c37-9480-dc43778ffc5f/next/gan/chainer-gan-lib-master/common/inception/inception_score.py", line 42, in inception_score
ims_batch = Variable(ims_batch, volatile=False)
File "/usr/local/lib/python3.4/dist-packages/chainer/variable.py", line 443, in init
kwargs, volatile='volatile argument is not supported anymore. '
File "/usr/local/lib/python3.4/dist-packages/chainer/utils/argument.py", line 4, in check_unexpected_kwargs
raise ValueError(message)
ValueError: volatile argument is not supported anymore. Use chainer.using_config

loss_gp calculation in DRAGAN updater

There appears to be an error in how the loss_gp term is computed. In the code snippet, x_perturb is used to calculate y_mid. However, y_mid is only used for the shape to create an array of ones, instead of the actual output from the discriminator. The values of x_perturb do not seem to be used to compute the loss_gp.

chainer-gan-lib/dragan/updater.py

Lines 40 to 43 in 18b7c7d

 y_mid = F.sigmoid(self.dis(x_perturb)) 

 dydx = self.dis.differentiable_backward(xp.ones_like(y_mid.data)) 

 dydx = F.sqrt(F.sum(dydx ** 2, axis=(1, 2, 3))) 

 loss_gp = self.lam * F.mean_squared_error(dydx, xp.ones_like(dydx.data))

inception module import error

after download inception module on common/inception/, I got following error

from inception.inception_score import inception_score, Inception
ModuleNotFoundError: No module named 'inception.inception_score'

do I edited evaluation.py.

from inception.inception_score

from common.inception.inception_score

is working on root directory of this repository and exec train.py is correct ?

float scale in F.average_pooling_2d

In updater of progressive GAN, the "reso", "reso_low" and "reso_high" at lines 41,47,48 must be converted to "int" as it will then be used as kernel and stride in F.average_pooling_2d

SN-DCGAN??

There is having SN-DCGAN's results.But I can't find the code.

why only consider GPU case?

hello, would you please also consider CPU(especially iDeep enbaled) case? In some files, we can see that you did "model.to_gpu()" or "chainer.cuda.to_cpu(**)" without checking directly. Thanks.

In PGGAN Why do unpooling successes image generation without using deconvolution?

Hello.
Although generally, in GAN I think that we use deconvolution, in PGGAN you use unpooling and convolution.
Where is the original thesis as for using unpooling and convolution?
Please tell me for that.
Thanks!

hope to implement CycleGAN, DiscoGAN, DualGAN

I look from introduction there is not implement in recently CycleGAN, DiscoGAN, DualGAN, Can support implement for them?

Does it support multiple GPU training?

how to reproduce inception / FID scores?

Which scores in the table are from your code? Do you have the commands to reproduce them exactly?

chainer-gan-lib/common/evaluation.py", line 142: 'numpy.ndarray' object has no attribute 'get'

Hello, I tried "python train.py --gpu -1 --algorithm dcgan --out result_dcgan " with chainerv5.0.0a1&iDeep, I saw following error:
File "/home/mingxiao/Documents/chainer-gan-lib/common/evaluation.py", line 175, in evaluation
mean, cov = get_mean_cov(model, np.asarray(xs).reshape((-1, 3, h, w)))
File "/home/mingxiao/Documents/chainer-gan-lib/common/evaluation.py", line 142, in get_mean_cov
mean = xp.mean(ys, axis=0).get()
AttributeError: 'numpy.ndarray' object has no attribute 'get'

Have you ever seen this before? Thanks in advance.

reading data from cifar-10-fid.npz file buggy

I downloaded your cifar-10-fid.npz file and when trying to load the data using fid = FID(stat["mean"], stat["cov"], mean, cov) then get following error: KeyError: 'mean is not a file in the archive'

Is there a discrepancy between the code and the file?

RuntimeError: cannot twice-differentiate an old style Function "sqrt"

Thanks for your interesting works.
I tried chainer-gan-lib/progressive, but it does not work well,

[Environments]
chainer-4.0.0a1, cupy-3.0.0a1, python 3.5.2, numpy 1.11.1, tensorflow 1.2.0.

The error message said, "RuntimeError: cannot twice-differentiate an old style Function "sqrt"".
It seems some trouble happened on twice-differentiate of backward().
However, I confirmed that chainer-gan-lib/wgan_gp, which also uses twice-differentiate, works for me.

All error messages are as follows:

/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/cupy/core/fusion.py:659: FutureWarning: cupy.core.fusion is experimental. The interface can change in the future.
  util.experimental('cupy.core.fusion')
Exception in main training loop: cannot twice-differentiate an old style Function "sqrt"
Traceback (most recent call last):
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/training/trainer.py", line 302, in run
    entry.extension(self)
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/contextlib.py", line 77, in __exit__
    self.gen.throw(type, value, traceback)
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/reporter.py", line 98, in scope
    yield
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/training/trainer.py", line 299, in run
    update()
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/training/updater.py", line 223, in update
    self.update_core()
  File "/project/nakamura-lab07/Work/seitaro-s/chainer-gan-lib/progressive/updater.py", line 85, in update_core
    loss_dis_total.backward()
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/variable.py", line 878, in backward
    self._backward_main(retain_grad)
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/contextlib.py", line 77, in __exit__
    self.gen.throw(type, value, traceback)
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/configuration.py", line 128, in using_config
    yield
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/variable.py", line 878, in backward
    self._backward_main(retain_grad)
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/variable.py", line 971, in _backward_main
    x._check_old_style_gradient()
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/variable.py", line 396, in _check_old_style_gradient
    self._old_style_grad_generator)
Will finalize trainer extensions and updater before reraising the exception.
Traceback (most recent call last):
  File "train.py", line 135, in <module>
    main()
  File "train.py", line 131, in main
    trainer.run()
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/training/trainer.py", line 313, in run
    six.reraise(*sys.exc_info())
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/six.py", line 693, in reraise
    raise value
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/training/trainer.py", line 302, in run
    entry.extension(self)
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/contextlib.py", line 77, in __exit__
    self.gen.throw(type, value, traceback)
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/reporter.py", line 98, in scope
    yield
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/training/trainer.py", line 299, in run
    update()
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/training/updater.py", line 223, in update
    self.update_core()
  File "/project/nakamura-lab07/Work/seitaro-s/chainer-gan-lib/progressive/updater.py", line 85, in update_core
    loss_dis_total.backward()
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/variable.py", line 878, in backward
    self._backward_main(retain_grad)
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/contextlib.py", line 77, in __exit__
    self.gen.throw(type, value, traceback)
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/configuration.py", line 128, in using_config
    yield
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/variable.py", line 878, in backward
    self._backward_main(retain_grad)
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/variable.py", line 971, in _backward_main
    x._check_old_style_gradient()
  File "/home/is/seitaro-s/.pyenv/versions/anaconda3-2.1.0/envs/chainerv4/lib/python3.5/site-packages/chainer/variable.py", line 396, in _check_old_style_gradient
    self._old_style_grad_generator)
RuntimeError: cannot twice-differentiate an old style Function "sqrt"

bug: Equalized Deconv layer

I figure self.inv_c = np.sqrt(2.0/(in_ch)) should be self.inv_c = np.sqrt(2.0/(in_ch*ksize**2)) like EqualizedConv2d layer. What do you think? :0

original code:

class EqualizedDeconv2d(chainer.Chain):
    def __init__(self, in_ch, out_ch, ksize, stride, pad):
        w = chainer.initializers.Normal(1.0) # equalized learning rate
        self.inv_c = np.sqrt(2.0/(in_ch))
        super(EqualizedDeconv2d, self).__init__()
        with self.init_scope():
            self.c = L.Deconvolution2D(in_ch, out_ch, ksize, stride, pad, initialW=w)
    def __call__(self, x):
        return self.c(self.inv_c * x)

TypeError: create_convolution_descriptor() got an unexpected keyword argument 'group'

I am not able to run training with progressive gan code.
After installing chainer from source according to instructions here and following the instructions for installation in this repo, I try to run progressive gan training code as follows:

python3 train.py

However, I get an exception, which reads as:

Exception in main training loop: create_convolution_descriptor() got an unexpected keyword argument 'group'
Traceback (most recent call last):
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/training/trainer.py", line 304, in run
    update()
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/training/updaters/standard_updater.py", line 136, in update
    self.update_core()
  File "/home/iaroslav/chainer-gan-lib/progressive/updater.py", line 59, in update_core
    y_real = self.dis(x_real, stage=self.stage)
  File "/home/iaroslav/chainer-gan-lib/progressive/net.py", line 180, in __call__
    h = F.leaky_relu(fromRGB(x))
  File "/home/iaroslav/chainer-gan-lib/progressive/net.py", line 28, in __call__
    return self.c(self.inv_c * x)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/links/connection/convolution_2d.py", line 164, in __call__
    group=self.group)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/functions/connection/convolution_2d.py", line 638, in convolution_2d
    y, = fnode.apply(args)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/function_node.py", line 257, in apply
    outputs = self.forward(in_data)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/function_node.py", line 364, in forward
    return self.forward_gpu(inputs)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/functions/connection/convolution_2d.py", line 167, in forward_gpu
    return self._forward_cudnn(x, W, b, y)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/functions/connection/convolution_2d.py", line 246, in _forward_cudnn
    group=self.group)
Will finalize trainer extensions and updater before reraising the exception.
Traceback (most recent call last):
  File "/home/iaroslav/.local/share/umake/ide/pycharm/helpers/pydev/pydevd.py", line 1596, in <module>
    globals = debugger.run(setup['file'], None, None, is_module)
  File "/home/iaroslav/.local/share/umake/ide/pycharm/helpers/pydev/pydevd.py", line 974, in run
    pydev_imports.execfile(file, globals, locals)  # execute the script
  File "/home/iaroslav/.local/share/umake/ide/pycharm/helpers/pydev/_pydev_imps/_pydev_execfile.py", line 18, in execfile
    exec(compile(contents+"\n", file, 'exec'), glob, loc)
  File "/home/iaroslav/chainer-gan-lib/progressive/train.py", line 166, in <module>
    main()
  File "/home/iaroslav/chainer-gan-lib/progressive/train.py", line 162, in main
    trainer.run()
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/training/trainer.py", line 318, in run
    six.reraise(*sys.exc_info())
  File "/usr/local/lib/python3.5/dist-packages/six.py", line 693, in reraise
    raise value
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/training/trainer.py", line 304, in run
    update()
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/training/updaters/standard_updater.py", line 136, in update
    self.update_core()
  File "/home/iaroslav/chainer-gan-lib/progressive/updater.py", line 59, in update_core
    y_real = self.dis(x_real, stage=self.stage)
  File "/home/iaroslav/chainer-gan-lib/progressive/net.py", line 180, in __call__
    h = F.leaky_relu(fromRGB(x))
  File "/home/iaroslav/chainer-gan-lib/progressive/net.py", line 28, in __call__
    return self.c(self.inv_c * x)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/links/connection/convolution_2d.py", line 164, in __call__
    group=self.group)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/functions/connection/convolution_2d.py", line 638, in convolution_2d
    y, = fnode.apply(args)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/function_node.py", line 257, in apply
    outputs = self.forward(in_data)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/function_node.py", line 364, in forward
    return self.forward_gpu(inputs)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/functions/connection/convolution_2d.py", line 167, in forward_gpu
    return self._forward_cudnn(x, W, b, y)
  File "/usr/local/lib/python3.5/dist-packages/chainer-4.0.0b2-py3.5.egg/chainer/functions/connection/convolution_2d.py", line 246, in _forward_cudnn
    group=self.group)
TypeError: create_convolution_descriptor() got an unexpected keyword argument 'group'

My OS: Ubuntu 16
I use python3
CuDNN version according to chainer.backends.cuda.cuda.cudnn.getVersion(): 7004
Cuda version according to nvcc --version: V8.0.61
cupy version: 3.0.0a1
numpy version: 1.14.0

Let me know if you need more information or if the issue should be posted in the chainer repo instead.

	y_mid = F.sigmoid(self.dis(x_perturb))
	dydx = self.dis.differentiable_backward(xp.ones_like(y_mid.data))
	dydx = F.sqrt(F.sum(dydx ** 2, axis=(1, 2, 3)))
	loss_gp = self.lam * F.mean_squared_error(dydx, xp.ones_like(dydx.data))