run hr_res('train') has error about tiny

[zhangjinsong3]

out of memory!
train to change the batchSize=4 ,for example : batchSzie =1.
Because the resnet101 need much more memory.

[tzhang2014]

thank you ! I have set opts.batchSize = 10 to opts.batchSize = 1in cnn_widerface.m and opts.batchSize = 256 to opts.batchSize = 1 in cnn_train_dag_hardmine.m ,but remain get the same error ,

[peiyunh]

Parameters set in hr_res101.m will override default ones in both cnn_widerface.m and cnn_train_dag_hardmine.m.

[FredaZhang338]

hi dear @peiyunh I haved compiled and tested as you said,and I can run the demo,but when I run hr_res101('train');I got the errors as floows:
( I have changed the MAX_INPUT_DIM in hr_res101.m to 1500)
Lab 1:
train: epoch 01: 1/269:
Lab 2:
train: epoch 01: 1/269:
Lab 3:
train: epoch 01: 1/269:
Lab 4:
train: epoch 01: 1/269:
Error using dagnn.Conv/forward (line 11)
Error detected on worker 1.
Out of memory on device. To view more detail about available memory on the GPU, use 'gpuDevice()'.
If the problem persists, reset the GPU by calling 'gpuDevice(1)'.

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

Error in dagnn.DagNN/forward (line 28)
obj.layers(l).block.forwardAdvanced(obj.layers(l)) ;

Error in cnn_train_dag_hardmine>process_epoch (line 286)
net.forward(inputs, opts.derOutputs);

Error in cnn_train_dag_hardmine>(spmd body) (line 158)
[stats_.train, prof_] = process_epoch(net_, state, opts, 'train') ;

Error in cnn_train_dag_hardmine>(spmd) (line 156)
spmd

Error in cnn_train_dag_hardmine (line 156)
spmd

Error in cnn_widerface (line 212)
[net, info] = trainFn(net, imdb, getBatchFn(batchGetter, opts, net.meta), ...

Error in hr_res101 (line 41)
cnn_widerface('inputSize', inputSize, ...

I have tried to reset the gpuDevice and tried to change the batchsize smaller ,but another error came out.So I am wondering if you have any idea to fix this problem. thank you so much!

[FredaZhang338]

I have fixed the probelm~thx