Closed rockywind closed 6 years ago
Hi!
Did you try by recompiling the MEX files in Matconvnet. You can do that by running:
vl_compilenn('EnableGpu',true);
Let me know if you solve the problem with that!
@ignaciorlando Thank you very much! I solved it following your advice,but I find another error that the GPU memory consumes a liitle,when I run the command "watch nviida-smi" training the model.
That is basically because the training set is not too big and the batch size is not too big either. In any case, the whole training process should not take more than half an hour on GPU.
@ignaciorlando I followed your suggestions running the command vl_compilenn('EnableGpu',true); but i got Could not find a valid NVCC executable\n. Any suggestions to me please
HI, My current configuration is: CUDA 8.0 Ubuntu16.04 Matlab2016b GPU:GTX1080
While running 'RedLesionDetection/scripts/script_train_cnn.m' in GPU mode I get the error: `train: epoch 01: 1/232:Error using vl_nnconv An input is not a numeric array (or GPU support not compiled).
Error in vl_simplenn (line 306) res(i+1).x = vl_nnconv(res(i).x, l.weights{1}, l.weights{2}, ...
Error in cnn_red_lesions_train>process_epoch (line 424) res = vl_simplenn(net, im, dzdy, res, ...
Error in cnn_red_lesions_train (line 150) [net,stats.train(epoch),prof] = process_epoch(net, state, opts, 'train') ;
Error in cnn_red_lesion_detection>train_cnn_from_scratch (line 93) [net, info] = cnn_red_lesions_train(net, imdb, getBatch(opts), ...
Error in cnn_red_lesion_detection (line 26) [net, info] = train_cnn_from_scratch(opts) ;
Error in script_train_cnn (line 12) [detector.net, info] = cnn_red_lesion_detection( ...` Thanks in advance!