search

vlfeat / matconvnet

MatConvNet: CNNs for MATLAB
Other
1.4k stars 753 forks source link

An unexpected error occurred during CUDA execution #730

Open FrancisYizhang opened 7 years ago

FrancisYizhang commented 7 years ago

Hi, I occurred a error when I run the _fast_rcnntrain.m. And my platform is Win7, matlab 2016a, CUDA7.5 and VS2012. The error is displayed followed.

Warning: An unexpected error occurred during CUDA execution. The CUDA error was: CUDA_ERROR_ILLEGAL_ADDRESS

In dagnn.Conv/backward (line 21) In dagnn.Layer/backwardAdvanced (line 120) In dagnn.DagNN/eval (line 117) In cnn_train_dag>processEpoch (line 217) In cnn_train_dag (line 87) In fast_rcnn_train (line 88)

Error: vl_nnconv An unexpected error occurred during CUDA execution. The CUDA error was: an illegal memory access was encountered

Can you help me? Thanks!

FrancisYizhang commented 7 years ago

My GPU is TITAN with 6 GB

Bhee commented 7 years ago

While running fast_rcnn_train.m. occured following error:

Warning:

An unexpected error occurred during CUDA execution. The CUDA error was: Unknown error code Error using vl_nnconv An unexpected error occurred during CUDA execution. The CUDA error was: unknown error

Error in dagnn.Conv/backward (line 21) [derInputs{1}, derParams{1}, derParams{2}] = vl_nnconv(...

Error in dagnn.Layer/backwardAdvanced (line 120) [derInputs, derParams] = obj.backward ...

Error in dagnn.DagNN/eval (line 119) obj.layers(l).block.backwardAdvanced(obj.layers(l)) ;

Error in cnn_train_dag>processEpoch (line 218) net.eval(inputs, params.derOutputs, 'holdOn', s < params.numSubBatches) ;

Error in cnn_train_dag (line 87) [net, state] = processEpoch(net, state, params, 'train') ;

Error in fast_rcnn_train (line 89) [net,info] = cnn_train_dag(net, imdb, @(i,b)

Spec : Windows 7, cuda toolkit 7.5,GPU: Titan X 12GB , Matlab 2014a, device driver version 375.70

Kindly help me to solve this.

Sarruedi commented 7 years ago

I currently experience the exact same error as you did. Did you ever find a solution for it?

guangdexuanwu commented 7 years ago

I'm doing a project based on fcn and I also meet this problem.It seems that the reason is the size of image.For example, if I change the size of one image whose size is (600,1000) to be smaller like(100,200), and restart Matlab (important), the problem will disappear. Maybe there are some bugs in MatConvNet.