search

hagaygarty / mdCNN

3D Convolutional Neural Network (CNN) for volumetric inputs. Matlab framework supporting 2D and 3D kernels
52 stars 40 forks source link

error with demomnist3d #4

Open shomedas opened 5 years ago

shomedas commented 5 years ago

When I run the demomnist3d I get error:

Finish preparing image 55000 of 60000 Finish preparing image 60000 of 60000 Finish image (test) 5000 of 10000 Finish image (test) 10000 of 10000 Dataset info - test: 10000, train: 60000, first sample size:=28 28 28, var=1576.90, min=0.000000, max=255.000000 Verifying backProp.. Error using - Matrix dimensions must agree.

Error in feedForward (line 86) net.layers{k}.outs.batchVar = mean(abs(input-net.layers{k}.outs.batchMean).^2,length(net.layers{k}.properties.sizeOut)+1) ;

Error in verifyBackProp (line 20) net = feedForward(net, input, 0);

Error in Train (line 39) verifyBackProp(net);

Error in demoMnist3D (line 12) net = Train(MNIST3d,net, length(MNIST3d.I)*2 ); % 2 epocs

shomedas commented 5 years ago

i am on ubuntu1.04, cuda 8.0

hagaygarty commented 5 years ago

Please use matlab 2016b and above. The error is due to an old matlab version

hagaygarty commented 5 years ago

the latest code is not using matlab arithmetic expands, so now older matlab versions are supported.