Code runs fine on CPU but gives ValueError on GPU

[shariqfarooq123]

I'm trying to build a Deconvolutional Network, model is built as follows:

`
model = Sequential([

Convolution2D(120,5,5,input_shape=(3,16,16),dim_ordering="th",activation="relu",border_mode="same"),

MaxPooling2D(pool_size=(2,2)),

Convolution2D(240,3,3,activation="relu",border_mode="same"),

MaxPooling2D(pool_size=(2, 2)),

Convolution2D(90,1,1,activation="relu",border_mode="same"),

MaxPooling2D(pool_size=(2, 2)),

Convolution2D(512,2,2,activation="relu"),

UpSampling2D(dim_ordering="th"),

Deconvolution2D(112,2,2,input_shape=(512,2,2),output_shape=(None,112,4,4),subsample=(2,2),dim_ordering="th",activation="relu",),

Deconvolution2D(64,2,2,input_shape=(112,4,4),output_shape=(None,64,8,8),subsample=(2,2),dim_ordering="th",activation="relu"),

Deconvolution2D(32,2,2,input_shape=(64,8,8),output_shape=(None,32,16,16),subsample=(2,2),dim_ordering="th",activation="relu"),

Deconvolution2D(4,1,1,input_shape=(32,16,16),output_shape=(None,4,16,16),dim_ordering="th",activation="relu"),

Flatten(),

])
`

If I set "optimizer=None" in .theanorc file i.e if code is run on CPU, No error occurs. But when run using GPU (optimizer=fast_run) following error occurs:
`

 Traceback (most recent call last):

File "C:/Users/Lab/PycharmProjects/srcovnet/shariq/cov2d.py", line 59, in <module>
model.fit(data_train,labels_train,nb_epoch=10)

File "C:\Users\Lab\Anaconda2\lib\site-packages\keras\models.py", line 652, in fit
sample_weight=sample_weight)

File "C:\Users\Lab\Anaconda2\lib\site-packages\keras\engine\training.py", line 1111, in fit
initial_epoch=initial_epoch)

File "C:\Users\Lab\Anaconda2\lib\site-packages\keras\engine\training.py", line 826, in _fit_loop
outs = f(ins_batch)

File "C:\Users\Lab\Anaconda2\lib\site-packages\keras\backend\theano_backend.py", line 811, in 
__call__
return self.function(*inputs)

File "C:\Users\Lab\Anaconda2\lib\site-packages\theano\compile\function_module.py", line 871, in 
__call__

storage_map=getattr(self.fn, 'storage_map', None))

File "C:\Users\Lab\Anaconda2\lib\site-packages\theano\gof\link.py", line 314, in raise_with_op
reraise(exc_type, exc_value, exc_trace)

File "C:\Users\Lab\Anaconda2\lib\site-packages\theano\compile\function_module.py", line 859, in 
__call__

outputs = self.fn()



ValueError: GpuDnnConv images and kernel must have the same stack size`

What is the problem?

[gvtulder]

Which version of Keras are you using? I looks similar to a problem I had before and that was fixed by #4631. If your problem is similar, upgrading to 1.2.0 might help. Keras 1.1.2 and older have an incorrect order of the filter shape in the deconvolution layer for Theano.

If this is indeed the cause of your problem, your code probably also doesn't do what you want in CPU mode. It's just that the GPU code has better checks. (This is fixed in the current Theano master and should be included in the next Theano release.)

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs, but feel free to re-open it if needed.