Some problem with running vl_simplenn.m

[huangsjb]

Hi, I have encountered some problem in trying to run vl_simplenn.m to test the model.

I simply run the sample code in the following tutorial:
http://www.vlfeat.org/matconvnet/pretrained/

And the following error message is printed:
" Reference to non-existent field 'filters'.

Error in vl_simplenn (line 153)
res(i+1).x = vl_nnconv(res(i).x, l.filters, l.biases, 'pad', l.pad, 'stride', l.stride) ;

Error in cnn_test (line 16)
res = vl_simplenn(net, im_) ; "

If someone has encountered the same problem as running the sample code? or have any experience on dealing with it?

Thanks a lot anyway.

[geyun12]

I meet the same problem with you... hope someone help us.

[huangsjb]

Also, I tried to do prediction using the cnn_mnist.m in the examples folder.

By changing the last layer to "softmax" (as discussed in #34), it works. But the tutorial example problem is not solved yet.

[geyun12]

If I used the imagenet-vgg-f model,then it runs successfully. But when I used the imagenet-vgg-verydeep-16, then I meet the same problem as yours.

[Jerrynet]

@huangsjb you can update matconvnet to the latest version (1.0-beta11), which supports legacy net structure.
They updated net weight structure from '.filters'/'.biases' to '.weights'
A quick workaround is change the code to this:

res(i+1).x = vl_nnconv(res(i).x, l.weights{1}, l.weights{2}, 'pad', l.pad, 'stride', l.stride) ;

Edit: beta12 comes with a 'legacy net' option, so no need to modify it.

[wangchuanxd]

@huangsjb Maybe your training network did not parallel for your training data setting. As you said ,you changed the layer to "softmax" and it worked.

[isalirezag]

I have the same issue.
Could anyone solve this problem?

I am just trying to use the pre-train model:

run vl_setupnn
net = load('imagenet-vgg-f.mat') ;

% obtain and preprocess an image
im = imread('lena.bmp') ;
im_ = single(im) ; % note: 255 range
im_ = imresize(im_, net.normalization.imageSize(1:2)) ;
im_ = im_ - net.normalization.averageImage(1) ;
% run the CNN
res = vl_simplenn(net,im_) ;

I received this error as well:

Error in vl_simplenn (line 178)
res(i+1).x = vl_nnconv(res(i).x, l.weights{1}, l.weights{2}, 'pad', l.pad, 'stride',
l.stride) ;

[Mona77]

I get the same error as well:

Attempt to execute SCRIPT vl_nnconv as a function:
/home/m/matconvnet-1.0-beta12/matlab/vl_nnconv.m

Error in vl_simplenn (line 178)
res(i+1).x = vl_nnconv(res(i).x, l.weights{1}, l.weights{2}, 'pad', l.pad, 'stride', l.stride) ;

Error in pre_train_vgg_s (line 15)
res = vl_simplenn(net, im_) ;

[wangchuanxd]

Sorry for replying your problem so late. I encountered the problem when I setted the error parameters beteween two layers. Then, I fixed the parameters and it worked.

wangchuansnnu@163.com

From: Mona Fathollahi
Date: 2015-05-28 12:53
To: vlfeat/matconvnet
CC: wangchuanxd
Subject: Re: [matconvnet] Some problem with running vl_simplenn.m (#131)
I get the same error as well:
Attempt to execute SCRIPT vl_nnconv as a function:
/home/m/matconvnet-1.0-beta12/matlab/vl_nnconv.m
Error in vl_simplenn (line 178)
res(i+1).x = vl_nnconv(res(i).x, l.weights{1}, l.weights{2}, 'pad', l.pad, 'stride', l.stride) ;
Error in pre_train_vgg_s (line 15)
res = vl_simplenn(net, im_) ;
—
Reply to this email directly or view it on GitHub.

[arjunrajanna]

I am trying to use it for cifar.
I get the same error as well. I am using the option "legacy net".

Thanks

[arjunrajanna]

Could some one help out?

[arjunrajanna]

Compiled the mex files with "vl_compilenn". That works.

[lenck]

Hi, Attempt to execute SCRIPT vl_nnconv as a function: is because the MatConvNet is not compiled. You should run vl_compilenn.
The error " Reference to non-existent field 'filters'. means that your network is of the old format. Use the vl_simplenn_tidy which would fix it.
Hope that this would help a bit! :)

[cora9298]

use 'net.meta.normalization' instead of 'net.normalization'

[sabirdvd]

I think this is a bug in the framework when you try to run more than 2 or 3 networks at same time, workspace, clear cmd doesn’t work either, only solution is to restart MATLAB and run the same lines again. What I did, I run the first network then I saved the weight then I restarted MATLAB, then I run the another one. I tried with this two laptops to confirm this bug.