why the fcnTrain.m running so slow? about matconvnet-fcn

[mjohn123]

Hello, I think that you did not active GPU mode. Find and change opts.train.gpus = [] to opts.train.gpus = 1 . It needs more than 5GB GPU

[brisker]

@mjohn123
In fcnTrain I haven't found opts.train.gpus = []
only like this:
% training options (SGD)
opts.train = struct([]) ;
[opts, varargin] = vl_argparse(opts, varargin) ;

trainOpts.batchSize = 20 ;

trainOpts.numSubBatches = 10 ;

trainOpts.continue = true ;
% trainOpts.gpus = [1,2,3,4] ;
% trainOpts.gpus = [1] ;
trainOpts.gpus = [2] ;
trainOpts.prefetch = true ;
trainOpts.expDir = opts.expDir ;
trainOpts.learningRate = 0.0001 * ones(1,50) ;

trainOpts.numEpochs = numel(trainOpts.learningRate) ;

[brisker]

@mjohn123 "opts.train.gpus = 1" or "opts.train.gpus = [1]" ??

[mjohn123]

Let read that issue and do it. It worked

#17