skaae / torch-gan Goto Github PK

Can you explain why you're using nn.gmodule instead of nn.sequential for the generator? It seems all the modules are sequential.

rzai@rzai00:~/prj/torch-gan/datasets$ python create_datasets.py
Traceback (most recent call last):
File "create_datasets.py", line 38, in
x = create_lfw()
File "create_datasets.py", line 27, in create_lfw
imgs = lfw_imgs(alignment='deepfunneled', size=64, crop=50)
File "create_datasets.py", line 16, in lfw_imgs
imgs, names_idx, names = lfw.LFW(alignment).arrays()
File "/home/rzai/prj/torch-gan/datasets/lfw/lfw.py", line 43, in init
self._install()
File "/home/rzai/prj/torch-gan/datasets/lfw/lfw.py", line 57, in _install
filepath = download(url, self.data_dir)
File "/home/rzai/prj/torch-gan/datasets/lfw/util.py", line 35, in download
urllib.urlretrieve(url, filepath)
File "/usr/lib/python2.7/urllib.py", line 94, in urlretrieve
return _urlopener.retrieve(url, filename, reporthook, data)
File "/usr/lib/python2.7/urllib.py", line 284, in retrieve
"of %i bytes" % (read, size), result)
urllib.ContentTooShortError: retrieval incomplete: got only 17610303 out of 108761145 bytes

rzai@rzai00:~/prj/torch-gan/datasets$

.....
Generator network:
nn.gModule
Copy model to gpu
/home/sdq/torch/install/bin/luajit: /home/sdq/.luarocks/share/lua/5.1/nn/THNN.lua:109: wrong number of arguments for function call
stack traceback:
[C]: in function 'v'
/home/sdq/.luarocks/share/lua/5.1/nn/THNN.lua:109: in function 'HardTanh_updateOutput'
/home/sdq/.luarocks/share/lua/5.1/nn/HardTanh.lua:17: in function 'forward'
train_lfw.lua:180: in function 'getSamples'
train_lfw.lua:192: in main chunk
[C]: in function 'dofile'
.../sdq/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670

Hi,

I can't make your create_dataset.py to work. One sanity check failing: np.array(self.imgs.tolist().getdata()).sum() seems to be zero.

The actual error I got is:

Traceback (most recent call last):
  File "create_datasets.py", line 38, in <module>
    x = create_lfw()
  File "create_datasets.py", line 27, in create_lfw
    imgs = lfw_imgs(alignment='deepfunneled', size=64, crop=50)
  File "create_datasets.py", line 19, in lfw_imgs
    img = img[crop:-crop, crop:-crop]
IndexError: too many indices for array

Thank you,
Jake

Looks interesting and impressive results after only few epochs. How would I go about training this on my own dataset of face photos, rather than the In the Wild dataset? From other posts it appears these need to all be cropped to same dimensions and aligned similarly. Is there any other gotchas? Pixel width / height? Any recommendations for minimum # of photos required to work properly?