Error message of vl_nnconv() when using GPU mode with OS X 10.9 about matconvnet

[lenck]

Hi, sorry for so late response... :P
This never happenned to me, but does other GPU related stuff works in Matlab? I.e. computation with gpu arrays or the Mathworks example here?
Then, the issue can be also that you are resuming training which was being run on CPU and then you want to resume for GPU - in the current implementation it is not checked and not autmoatically transferred... So, maybe when the error occurs, check whether all the first three arguments in vl_nnconv are gpuArrays...
Sorry, it's more a list of random ideas, but that would be what I would try to do...

[luzhi]

@lenck Thanks for your reply. I think I find the solution to the problem, which is the compatibility of CUDA version and the Matlab version. If I installed CUDA 5.5 with Matlab 2014a, it works.

[rohitdurvasula]

Hi,
I have CUDA 6.4, Ubuntu 14.04 with Matlab R2014b. I also get the same error. The same function vl_nnconv works in CPU mode. If I gather the arrays and perform vl_nnconv it works. So I guess the problem has to be with the GPU implementation of the code.

Any suggestions without downgrading to 5.5? Without this imagenet example is impossible to run..!

[khouloud123]

Hi,
When i run the exercice 5 from VGG Convolutional Neural Networks Practical some erreus appears
such as :
Error using load
Can't read file C:\Users\MONDE
INFO\Desktop\CNNstart\practical-cnn-2015a\practical-cnn-2015a\data\imagenet-vgg-verydeep-16.mat.

why ?

THANKS