search

martinkersner / train-DeepLab

Train DeepLab for Semantic Image Segmentation
MIT License
172 stars 76 forks source link

How to solve the problem: Unexpected label *** #31

Open zhangrui0828 opened 7 years ago

zhangrui0828 commented 7 years ago

Hi Martin, I found your tutorial very help and I am following it to train a model step by step. The DeepLab model : DeepLab_LargeFOV. Dataset: original PASCAL VOC12. Init model is vgg16_20M. Solver: exper/voc12/config/DeepLab_LargeFOV/solver_train.prototxt train_net: exper/voc12/config/DeepLab_LargeFOV/train_train.prototxt source: exper/voc12/list/train.txt

When I run run_pascal_strong.sh, an error always occur and I can't further proceed. The error is shown below. …... I0612 15:44:39.689864 12910 net.cpp:208] This network produces output accuracy I0612 15:44:39.689891 12910 net.cpp:467] Collecting Learning Rate and Weight Decay. I0612 15:44:39.689899 12910 net.cpp:219] Network initialization done. I0612 15:44:39.689903 12910 net.cpp:220] Memory required for data: 3663953152 I0612 15:44:39.689975 12910 solver.cpp:41] Solver scaffolding done. I0612 15:44:39.689981 12910 solver.cpp:160] Solving DeepLab-LargeFOV I0612 15:44:39.689985 12910 solver.cpp:161] Learning Rate Policy: step F0612 15:44:41.198884 12910 softmax_loss_layer.cpp:86] Unexpected label 19 Check failure stack trace: @ 0x7efefd561daa (unknown) @ 0x7efefd561ce4 (unknown) @ 0x7efefd5616e6 (unknown) @ 0x7efefd564687 (unknown) @ 0x5482a8 caffe::SoftmaxWithLossLayer<>::Forward_cpu() @ 0x47413a caffe::Net<>::ForwardFromTo() @ 0x4743cf caffe::Net<>::ForwardPrefilled() @ 0x57a320 caffe::Solver<>::Solve() @ 0x419be8 train() @ 0x412278 main @ 0x7efefa399f45 (unknown) @ 0x417927 (unknown) @ (nil) (unknown) Aborted

The number after “Unpexted label” is always different at every time. Do you have any ideas on this? That would be really greatful. Thanks!

debayan commented 6 years ago

Your image needs to be single channel and it should not have a larger number in the pixel value than the number of classes you have specified in your softmax layer.