TensorFlow doesn't support NCHW

[hao-zi]

`UnimplementedError (see above for traceback):  Generic conv implementation only supports NHWC tensor format for now.
	 [[Node: Generator.1.1/conv1d/Conv2D = Conv2D[T=DT_FLOAT, data_format="NCHW", padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](Generator.1.1/conv1d/ExpandDims, Generator.1.1/conv1d/ExpandDims_1)]]`

my tensorflow is 1.4.0
python is 2.7

[hao-zi]

when i change NCHW to NHWC in tflib/ops/conv1d.py, i get errors:

Traceback (most recent call last):
  File "sample.py", line 79, in <module>
    fake_inputs = models.Generator(args.batch_size, args.seq_length, args.layer_dim, len(charmap))
  File "/home/tf/passgam/PassGAN-master/models.py", line 19, in Generator
    output = ResBlock('Generator.1', output, layer_dim)
  File "/home/tf/passgam/PassGAN-master/models.py", line 9, in ResBlock
    output = lib.ops.conv1d.Conv1D(name+'.1', dim, dim, 5, output)
  File "/home/tf/passgam/PassGAN-master/tflib/ops/conv1d.py", line 93, in Conv1D
    data_format='NHWC'
  File "/home/tf/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/nn_ops.py", line 2204, in conv1d
    data_format=data_format)
  File "/home/tf/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/gen_nn_ops.py", line 631, in conv2d
    data_format=data_format, name=name)
  File "/home/tf/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
    op_def=op_def)
  File "/home/tf/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2958, in create_op
    set_shapes_for_outputs(ret)
  File "/home/tf/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2209, in set_shapes_for_outputs
    shapes = shape_func(op)
  File "/home/tf/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2159, in call_with_requiring
    return call_cpp_shape_fn(op, require_shape_fn=True)
  File "/home/tf/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 627, in call_cpp_shape_fn
    require_shape_fn)
  File "/home/tf/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 691, in _call_cpp_shape_fn_impl
    raise ValueError(err.message)
ValueError: Dimensions must be equal, but are 10 and 128 for 'Generator.1.1/conv1d/Conv2D' (op: 'Conv2D') with input shapes: [1024,1,128,10], [1,5,128,128].

[brannondorsey]

It looks like you are running this on a CPU. If I remember correctly, I've run into this issue with Tensorflow CNNs run on CPUs before. I've only tested this code on CUDA supported GPU devices. I don't have the bandwidth to dig into CPU support right now, especially given how slow the performance will likely be, but likely others have posted on this or a similar issue w/ tensorflow conv2D before.

Good luck!

[mikeryan]

I'm having the same problem. How do you get the code to run on a GPU? I installed tensorflow using pip and my system has several nVidia 1080 GPUs, but the code does not appear to use them.

[brannondorsey]

@mikeryan, this should do the trick:

pip Install tensorflow-gpu

I'll add that to the requirements.txt file and upload it tonight.

[mikeryan]

After installing tensorflow-gpu using PIP, I had to install the following packages using apt-get install on Debian Squeeze (9.0): libcublas8.0 libcusolver8.0 libcudart8.0 libcufft8.0 libcurand8.0 Additionally I had to join the NVIDIA developer program and install "cuDNN v6.0 Runtime Library for Ubuntu16.04 (Deb)" from https://developer.nvidia.com/rdp/cudnn-download After doing the above, it runs well.

[iBM88]

I have the same problem. The packages mikeryan mentioned are already installed on my machine with version 7.5. I installed cuDNN too but it didn't help. Any ideas?