Interactive Image Generation via Generative Adversarial Networks

Thank you for your excellent work. I believe this code will be more readable and extensible if in PyTorch. So will the PyTorch version be released? A link of third-party implementation is also fine, though I cannot find one for now. Specifically, I'm interested in latent vector prediction and projection section.

When I run this python, THEANO_FLAGS='device=cpu, floatX=float32, nvcc.fastmath=True' python iGAN_main.py --model_name outdoor_64

I got this error.

Traceback (most recent call last): File "iGAN_main.py", line 40, in model_class = locate('model_def.%s' % args.model_type) File "/usr/lib/python2.7/pydoc.py", line 1521, in locate nextmodule = safeimport(join(parts[:n+1], '.'), forceload) File "/usr/lib/python2.7/pydoc.py", line 342, in safeimport raise ErrorDuringImport(path, sys.exc_info()) pydoc.ErrorDuringImport: problem in model_def.dcgan_theano - <type 'exceptions.ImportError'>: No module named cuda.dnn

I already installed CUDA and cuDNN. How can I run this code? Thank you.

When I run this script, the following error happens: File "train_predict_z.py", line 19, in from lib import AlexNet File "../lib/AlexNet.py", line 1, in from lasagne.layers import InputLayer, Conv2DLayer File "/home/atlantix/.local/lib/python2.7/site-packages/lasagne/init.py", line 19, in from . import layers File "/home/atlantix/.local/lib/python2.7/site-packages/lasagne/layers/init.py", line 7, in from .pool import * File "/home/atlantix/.local/lib/python2.7/site-packages/lasagne/layers/pool.py", line 6, in from theano.tensor.signal import downsample ImportError: cannot import name downsample

I'm sure the version of theano and lasagne is the latest. And I have tried to reinstall lasagne but the error is still there. Is lasagne not working? Thanks.

Is the model for projecting an image back onto the manifold also packaged with the pre-trained models listed in the README? The code is organized very well and I've been able to use the Model class in ``dcgan_theano.pyto sample from the downloaded models, but I'm less clear how to do the inverse mappingx => z` through the projection model and optimization objective.

Ubuntu16.04 GTX960 2G

Traceback (most recent call last): File "iGAN_main.py", line 38, in model_G = model_class.Model(model_name=args.model_name, model_file=args.model_file) File "/home/tzatter/iGAN/model_def/dcgan_theano.py", line 24, in init self._gen = self.def_gen(self.gen_params, self.gen_pl, self.n_layers, self.n_f) File "/home/tzatter/iGAN/model_def/dcgan_theano.py", line 34, in def_gen gx = gen_test(z, gen_params, gen_pl, n_layers=n_layers, n_f=n_f) File "/home/tzatter/iGAN/model_def/dcgan_theano.py", line 319, in gen_test hout = relu(batchnorm(deconv(hin, w, subsample=(2, 2), border_mode=(2, 2)), u=u, s=s, g=g, b=b)) File "/home/tzatter/iGAN/lib/ops.py", line 90, in deconv img = gpu_contiguous(X) File "/usr/local/lib/python2.7/dist-packages/theano/gof/op.py", line 602, in call node = self.make_node(_inputs, *_kwargs) File "/usr/local/lib/python2.7/dist-packages/theano/sandbox/cuda/basic_ops.py", line 3963, in make_node input = as_cuda_ndarray_variable(input) File "/usr/local/lib/python2.7/dist-packages/theano/sandbox/cuda/basic_ops.py", line 46, in as_cuda_ndarray_variable return gpu_from_host(tensor_x) File "/usr/local/lib/python2.7/dist-packages/theano/gof/op.py", line 602, in call node = self.make_node(_inputs, *_kwargs) File "/usr/local/lib/python2.7/dist-packages/theano/sandbox/cuda/basic_ops.py", line 139, in make_node dtype=x.dtype)()]) File "/usr/local/lib/python2.7/dist-packages/theano/sandbox/cuda/type.py", line 95, in init (self.class.name, dtype, name)) TypeError: CudaNdarrayType only supports dtype float32 for now. Tried using dtype float64 for variable None

Jun-Yan Zhu, thank you for the awesome code.

I am trying to create my own dataset. I created the hdf5 file using the given script on the Readme, and now I want to launch the train_dcgan.py file on the hdf5 file. I also defined the model parameters in train_dcgan_config.py.

This is what I get after computing (hdf5 file's name is mickey_64) :

[model_name] = mickey_64 [ext] = [data_file] = ../datasets/mickey_64.hdf5 [cache_dir] = ./cache/mickey_64/ [batch_size] = 128 [update_k] = 2 [save_freq] = 1 [lr] = 0.0002 [weight_decay] = 1e-05 [b1] = 0.5 ./cache/mickey_64/web_dcgan/images LOADING DATASET... name = ../datasets/mickey_64.hdf5, ntrain = 884, ntest = 46 0.01 secs to load data Traceback (most recent call last): File "train_dcgan.py", line 66, in vis_idxs = py_rng.sample(np.arange(len(test_x)), n_vis) File "/home/aeon7/anaconda3/lib/python3.6/random.py", line 313, in sample raise TypeError("Population must be a sequence or set. For dicts, use list(d).") TypeError: Population must be a sequence or set. For dicts, use list(d).

I don't understand where this error comes from.

This is great stuff!

The README.md mentioned that there is a plan to support variational autoencoder, which I very much look forward to. My understanding is that the vanilla VAE tends to create blurry images (perhaps due to the disentangled representation that it learns), and I wonder if the planned VAE feature will be based on the vanilla VAE, or something like the autoencoding beyond pixels, or adversarial autoencoders?

Device Information

ThinkPad R400 with ArchLinux my GPU is AMD rather than Nvidia.

Problem

when I exec

THEANO_FLAGS='device=gpu0, floatX=float32, nvcc.fastmath=True' python iGAN_main.py --model_name outdoor_64'

It says:

Traceback (most recent call last):
  File "iGAN_main.py", line 9, in <module>
    import constrained_opt
  File "/home/konjac/code/iGAN/constrained_opt.py", line 3, in <module>
    from lib.rng import np_rng
  File "/home/konjac/code/iGAN/lib/rng.py", line 2, in <module>
    from theano.sandbox.rng_mrg import MRG_RandomStreams as RandomStreams
  File "/usr/lib/python3.6/site-packages/theano/__init__.py", line 67, in <module>
    from theano.configdefaults import config
  File "/usr/lib/python3.6/site-packages/theano/configdefaults.py", line 113, in <module>
    in_c_key=False)
  File "/usr/lib/python3.6/site-packages/theano/configparser.py", line 285, in AddConfigVar
    configparam.__get__(root, type(root), delete_key=True)
  File "/usr/lib/python3.6/site-packages/theano/configparser.py", line 333, in __get__
    self.__set__(cls, val_str)
  File "/usr/lib/python3.6/site-packages/theano/configparser.py", line 344, in __set__
    self.val = self.filter(val)
  File "/usr/lib/python3.6/site-packages/theano/configdefaults.py", line 100, in filter
    % (self.default, val, self.fullname)))
ValueError: Invalid value ("cpu") for configuration variable "gpu0". Valid options start with one of "device", "opencl", "cuda"

What should I do to solve this? Is it caused because of my gpu brand? So is it possible to run this app on amd gpu?

If you are interested in image editing and GANs, I would like to share with you our recent project CycleGAN for learning an image-to-image translation (i.e. pix2pix) without input-output pairs. Here are a few applications of the algorithm.

Hello professor, I saw in an issue a few years ago that you mentioned that there would be a PyTorch version of this code, but a few days ago you replied that there was an iGAN improvement work, do you mean that there will be no PyTorch version of this code? I am very interested in iGAN and want to use it to do some work to complete my study, but Theano has stopped updating. It is not only difficult to read the code, but also incompatible with many versions when reproducing the code. If the PyTorch version of iGAN will be released in the near future, it will save a lot of time for my follow-up work. Thanks！

Theano1.0 has changed the interface of the function-

The original interface was: class GpuAllocEmpty(GpuOp); but now: class GpuAllocEmpty(HideC, AllocEmpty);

What should I do if I want to change to the corresponding interface？ Thank you.

hi, i'm running this command after I have installed everything

iGAN2noui A$ THEANO_FLAGS='device=gpu0, floatX=float32, nvcc.fastmath=True' python iGAN_main.py --model_name outdoor_64

and get this error. everything is installed as you described. i tried it on macos and ubuntu. same error.

Traceback (most recent call last):
  File "iGAN_main.py", line 4, in <module>
    import qdarkstyle
ImportError: No module named qdarkstyle

What would you recommend for someone who's running this code now? What should I install? I am asking this because theano backend has been changed and on running the given code now, I am getting numerous errors related to the same.

Do give some suggestions in this matter.

Traceback (most recent call last): File "generate_samples.py", line 32, in model_class = locate('model_def.%s' % args.model_type) File "/usr/lib/python2.7/pydoc.py", line 1504, in locate nextmodule = safeimport(join(parts[:n+1], '.'), forceload) File "/usr/lib/python2.7/pydoc.py", line 341, in safeimport raise ErrorDuringImport(path, sys.exc_info()) pydoc.ErrorDuringImport: problem in model_def.dcgan_theano - <type 'exceptions.ImportError'>: cannot import name raise_from

This is the error, I am getting. It would be great if you could help me out here.