I'm trying to implement traditional 2d convolution on a GPU.
input_0 is of shape [number of batches, num_of_channels, height, width]
filter is of shape [3]
filter1 = theano.shared(np.asarray([.5, 1, .5],
dtype=theano.config.floatX), borrow=True)
img_shp = (input_shape[0]*input_shape[1], input_shape[2], input_shape[3])
input_v = theano.tensor.signal.conv.conv2d(input_0.reshape(img_shp),
filter.reshape((1,3)),
border_mode='full',
image_shape=img_shp,
filter_shape=(1, 1, 3))[:,:,1:-1]
Is there a similar 2d convolution function for GPUs? If not, how can I
rewrite the code so that with theano.tensor.nnet.conv2d so that I can use
dnn_conv?
Thanks,
James
--
---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
