No variable of name 'label' could be found in the DAG.

[samadak]

Hello everyone
I want to finetune vgg16 by using my own dataset.It has 32 classes but I faced with some error.
train: epoch 01: 1/ 62:148 im2= imdb.images.data(:,:,:,batch);
Error using dagnn.DagNN/eval (line 83)
No variable of name 'label' could be found in the DAG.

Error in cnn_train_dag>processEpoch (line 253)
net.eval(inputs, params.derOutputs, 'holdOn', s < params.numSubBatches) ;

Error in cnn_train_dag (line 105)
[net, state] = processEpoch(net, state, params, 'train') ;

Error in fine_test0 (line 113)
info = cnn_train_dag(net, imdb, @(i,b) getBatch(bopts,i,b), opts.train, 'val', find(imdb.images.set == 3)) ;
%and this my code:
%finetune vgg16 for cifar10 dataset
function [net, info] = vgg_train(imdb, expDir)
% Demonstrated MatConNet on CIFAR-10 using DAG
% run(fullfile(fileparts(mfilename('fullpath')), '../../', 'matlab', 'vl_setupnn.m')) ;
run matlab/vl_setupnn
% imdb=load('imdbs.mat', 'imdb')
% imdb=load('imdb_cifar10.mat')
load('imdbs.mat', 'imdb')
% some common options
opts.train.batchSize = 100;
opts.train.numEpochs = 20 ;
opts.train.continue = true ;
opts.train.gpus = [];%[1] ;
opts.train.learningRate = [1e-1ones(1, 10), 1e-2ones(1, 5)];
opts.train.weightDecay = 3e-4;
opts.train.momentum = 0.;
% opts.train.expDir = expDir;
bopts.useGpu = numel(opts.train.gpus) > 0 ; % Usually keep at 0, seems to only work with 3D data.
opts.train.numSubBatches = 1;
% getBatch options
bopts.useGpu = numel(opts.train.gpus) > 0 ;

% network definition!
% MATLAB handle, passed by reference
net = dagnn.DagNN() ;
%vgg16 artitectur
net.addLayer('conv1_1', dagnn.Conv('size', [3 3 3 64], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]), {'input'},{'conv1'}, {'conv1_1f' 'conv1_1b'}); 

net.addLayer('relu1_1', dagnn.ReLU(), {'conv1'}, {'relu1'}, {});
net.addLayer('conv1_2', dagnn.Conv('size', [3 3 64 64], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]),{'relu1'}, {'conv2'}, {'conv1_2f' 'conv1_2b'});

net.addLayer('relu1_2', dagnn.ReLU(), {'conv2'}, {'relu2'}, {});
%net.addLayer('lrn1', dagnn.LRN('param', [5 1 0.0001/5 0.75]), {'relu1'}, {'lrn1'}, {});
net.addLayer('pool1', dagnn.Pooling('method', 'max', 'poolSize', [2, 2], 'stride', [2 2], 'pad', [0 0 0 0]),{'relu2'}, {'pool1'}, {});
net.addLayer('conv2_1', dagnn.Conv('size', [3 3 64 128], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]),{'pool1'}, {'conv3'}, {'conv2_1f' 'conv2_1b'});

net.addLayer('relu2_1', dagnn.ReLU(), {'conv3'}, {'relu3'}, {});
%layer8
net.addLayer('conv2_2', dagnn.Conv('size', [3 3 128 128], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]), {'relu3'}, {'conv4'}, {'conv2_2f' 'conv2_2b'});
net.addLayer('relu2_2', dagnn.ReLU(), {'conv4'}, {'relu4'}, {});
%net.addLayer('lrn2', dagnn.LRN('param', [5 1 0.0001/5 0.75]), {'relu2'}, {'lrn2'}, {});
%change padding in layer10
net.addLayer('pool2', dagnn.Pooling('method', 'max', 'poolSize', [2, 2], 'stride', [2 2], 'pad', [0 0 0 0]),{'relu4'}, {'pool2'}, {});

net.addLayer('conv3_1', dagnn.Conv('size', [3 3 128 256], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]),{'pool2'}, {'conv5'}, {'conv3_1f' 'conv3_1b'});

net.addLayer('relu3_1', dagnn.ReLU(), {'conv5'}, {'relu5'}, {});
net.addLayer('conv3_2', dagnn.Conv('size', [3 3 256 256], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]),{'relu5'}, {'conv6'}, {'conv3_2f' 'conv3_2b'});

net.addLayer('relu3_2', dagnn.ReLU(), {'conv6'}, {'relu6'}, {});
%layer 15 ok
net.addLayer('conv3_3', dagnn.Conv('size', [3 3 256 256], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]), {'relu6'}, {'conv7'}, {'conv3_3f' 'conv3_3b'});

net.addLayer('relu3_3', dagnn.ReLU(), {'conv7'}, {'relu7'}, {});
%net.addLayer('lrn3', dagnn.LRN('param', [5 1 0.0001/5 0.75]), {'relu2'}, {'lrn2'}, {});
%change in padding in 17 layer
net.addLayer('pool3', dagnn.Pooling('method', 'max', 'poolSize', [2, 2], 'stride', [2 2], 'pad', [0 0 0 0]),{'relu7'}, {'pool3'}, {});

%%%55
net.addLayer('conv4_1', dagnn.Conv('size', [3 3 256 512], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]),{'pool3'}, {'conv8'}, {'conv4_1f' 'conv4_1b'});

net.addLayer('relu4_1', dagnn.ReLU(), {'conv8'}, {'relu8'}, {});
net.addLayer('conv4_2', dagnn.Conv('size', [3 3 512 512], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]), {'relu8'}, {'conv9'}, {'conv4_2f' 'conv4_2b'});

net.addLayer('relu4_2', dagnn.ReLU(), {'conv9'}, {'relu9'}, {});
net.addLayer('conv4_3', dagnn.Conv('size', [3 3 512 512], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]), {'relu9'}, {'conv10'}, {'conv4_3f' 'conv4_3b'});

net.addLayer('relu4_3', dagnn.ReLU(), {'conv10'}, {'relu10'}, {});
%net.addLayer('lrn4', dagnn.LRN('param', [5 1 0.0001/5 0.75]), {'relu2'}, {'lrn2'}, {});
%layer24 change in padding
net.addLayer('pool4', dagnn.Pooling('method', 'max', 'poolSize', [2, 2], 'stride', [2 2], 'pad', [0 0 0 0]),{'relu10'}, {'pool4'}, {});

net.addLayer('conv5_1', dagnn.Conv('size', [3 3 512 512], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]),{'pool4'}, {'conv11'}, {'conv5_1f' 'conv5_1b'});

net.addLayer('relu5_1', dagnn.ReLU(), {'conv11'}, {'relu11'}, {});
net.addLayer('conv5_2', dagnn.Conv('size', [3 3 512 512], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]), {'relu11'}, {'conv12'}, {'conv5_2f' 'conv5_2b'});

net.addLayer('relu5_2', dagnn.ReLU(), {'conv12'}, {'relu12'}, {});
%29 layer
net.addLayer('conv5_3', dagnn.Conv('size', [3 3 512 512], 'hasBias', true, 'stride', [1, 1], 'pad', [1 1 1 1]), {'relu12'}, {'conv13'}, {'conv5_3f' 'conv5_3b'});

net.addLayer('relu5_3', dagnn.ReLU(), {'conv13'}, {'relu13'}, {});
net.addLayer('pool5', dagnn.Pooling('method', 'max', 'poolSize', [2, 2], 'stride', [2 2], 'pad', [0 0 0 0]), {'relu13'}, {'pool5'}, {});

net.addLayer('conv6', dagnn.Conv('size', [7 7 512 4096], 'hasBias', true, 'stride', [1, 1], 'pad', [0 0 0 0]), {'pool5'}, {'conv14'}, {'conv6f' 'conv6b'});

net.addLayer('relu6', dagnn.ReLU(), {'conv14'}, {'relu14'}, {});
net.addLayer('drop6', dagnn.DropOut('rate', 0.5), {'relu14'}, {'drop6'}, {});
net.addLayer('conv7', dagnn.Conv('size', [1 1 4096 4096], 'hasBias', true, 'stride', [1, 1], 'pad', [0 0 0 0]), {'drop6'}, {'conv15'}, {'conv7f' 'conv7b'});

net.addLayer('relu7', dagnn.ReLU(), {'conv15'}, {'relu15'}, {});
net.addLayer('drop7', dagnn.DropOut('rate', 0.5), {'relu15'}, {'drop7'}, {});
%%%chane in class number
net.addLayer('classifier', dagnn.Conv('size', [1 1 4096 32], 'hasBias', true, 'stride', [1, 1], 'pad', [0 0 0 0]), {'drop7'}, {'classifier'}, {'conv8f' 'conv8b'});

net.addLayer('prob', dagnn.SoftMax(), {'classifier'}, {'prob'}, {});
% net.addLayer('objective', dagnn.Loss('loss', 'log'), {'prob', 'label'}, {'objective'}, {});
% net.addLayer('error', dagnn.Loss('loss', 'classerror'), {'prob','label'}, 'error') ;
%
%
% initialization of the weights (CRITICAL!!!!)
% if(numel(varargin) > 0)
% initNet_FineTuning(net, netPre);
% else
% initNet_He(net);
% end
% %train
% iinitNet(net, 1/100);

% do the training!
info = cnn_train_dag(net, imdb, @(i,b) getBatch(bopts,i,b), opts.train, 'val', find(imdb.images.set == 3)) ;    

end

function initNet(net, f)
net.initParams();
%

f_ind = net.layers(1).paramIndexes(1);
b_ind = net.layers(1).paramIndexes(2);
net.params(f_ind).value = 10*f*randn(size(net.params(f_ind).value), 'single');
net.params(f_ind).learningRate = 1;
net.params(f_ind).weightDecay = 1;

for l=2:length(net.layers)
	% is a convolution layer?
	if(strcmp(class(net.layers(l).block), 'dagnn.Conv'))
		f_ind = net.layers(l).paramIndexes(1);
		b_ind = net.layers(l).paramIndexes(2);

		[h,w,in,out] = size(net.params(f_ind).value);
		net.params(f_ind).value = f*randn(size(net.params(f_ind).value), 'single');
		net.params(f_ind).learningRate = 1;
		net.params(f_ind).weightDecay = 1;

		net.params(b_ind).value = f*randn(size(net.params(b_ind).value), 'single');
		net.params(b_ind).learningRate = 0.5;
		net.params(b_ind).weightDecay = 1;
	end
end

end
% getBatch for IMDBs that are too big to be in RAM
function inputs = getBatch(opts, imdb, batch)
%images = imdb.images.data(:,:,:,batch) ;
labels = imdb.images.labels(1,batch) ;
% labels(labels>10)=10;
im2= imdb.images.data(:,:,:,batch);
im2=im2(1:224,1:224,:,:);
s=size(im2);

images=zeros(s(1),s(2),3,s(4),'single');
images(:,:,1,:)=im2;
images(:,:,2,:)=im2;
images(:,:,3,:)=im2;
if opts.useGpu > 0
	images = gpuArray(images) ;
end

inputs = {'input', images, 'label', labels} ;

end