Matlab - CuDNN Error with MatConvNet

Question

I am trying to use Matlab and Matconvnet for Deep Learning and particularly to work on this implementation : Finding Tiny Faces .

I am working on Ubuntu 16.04, with Cuda 9 (GTX 1080Ti). For the implementation, I am using Matlab R2017b. I have follow the instructions to install and compile Matconvnet vl_compilenn('enableImreadJpeg', true) and I have passed the furnished test.

When I try to run the first function bboxes = tiny_face_detector('data/demo/selfie.jpg', './selfie.png', 0.5, 0.1, 1) I have an error with CuDNN :

Error using vl_nnconv
forward: cuDNN error [cudnn:
"/home/alexattia/Work/RecVis/tiny/matconvnet/matlab/src/bits/impl/nnconv_cudnn.cu":141
(CUDNN_STATUS_BAD_PARAM)]

Error in dagnn.Conv/forward (line 11)
      outputs{1} = vl_nnconv(...

Error in dagnn.Layer/forwardAdvanced (line 85)
      outputs = obj.forward(inputs, {net.params(par).value}) ;

Error in dagnn.DagNN/eval (line 91)
  obj.layers(l).block.forwardAdvanced(obj.layers(l)) ;

Error in tiny_face_detector (line 130)
  net.eval(inputs);

Do you have any idea how to solve this ?
Thank you very much in advance.

Answer 1

Most probably, your CPU setting is still on. If you have changed that to GPU, the images and and params of the network need to be of type gpuArray.