getting error during training about hidden-two-stream HOT 7 CLOSED

Hi, the error message is clear. You need to make a directory named "logs_end" in your current working directory to save models.

The "logs_end" param is set in solver.prototxt. Hope this helps.

from hidden-two-stream.

Hi, Thank you so much for the help that worked for me. but unfortunately i end up with another error. I created one more directory in CD and given path to train_rgb_split1.txt and val_rgb_split1.txt. It is able to open train_rgb_split1.txt but could not load.

E0514 10:35:32.695055 4299 io.cpp:453] Could not load file /home/adi/Hidden-Two-Stream/models/ucf101_split1_unsup_end/new/WallPushups/v_WallPushups_g21_c06/image_0071.jpg
F0514 10:35:32.695204 4299 multi_frame_data_layer.cpp:65] Check failed: ReadSegmentMultiRGBToDatum(lines_[lines_id_].first, lines_[lines_id_].second, offsets, new_height, new_width, new_length, &datum)
*** Check failure stack trace: ***
@ 0x7f45dbd985cd google::LogMessage::Fail()
@ 0x7f45dbd9a433 google::LogMessage::SendToLog()
@ 0x7f45dbd9815b google::LogMessage::Flush()
@ 0x7f45dbd9ae1e google::LogMessageFatal::~LogMessageFatal()
@ 0x7f45dc6606b9 caffe::MultiFrameDataLayer<>::DataLayerSetUp()
@ 0x7f45dc5f6ae3 caffe::BasePrefetchingDataLayer<>::LayerSetUp()
@ 0x7f45dc6b7902 caffe::Net<>::Init()
@ 0x7f45dc6b9121 caffe::Net<>::Net()
@ 0x7f45dc6c139a caffe::Solver<>::InitTrainNet()
@ 0x7f45dc6c26d7 caffe::Solver<>::Init()
@ 0x7f45dc6c2a7a caffe::Solver<>::Solver()
@ 0x7f45dc44bca3 caffe::Creator_SGDSolver<>()
@ 0x40a6e8 train()
@ 0x4075a8 main
@ 0x7f45da84c830 __libc_start_main
@ 0x407d19 _start
@ (nil) (unknown)
Aborted (core dumped)

from hidden-two-stream.

Hi, according to the error, could you check that if there is indeed an image file in that location (/home/adi/Hidden-Two-Stream/models/ucf101_split1_unsup_end/new/WallPushups/v_WallPushups_g21_c06/image_0071.jpg)? Because it seems that the code couldn't find it.

One possibility is, we use different video codec to decode the videos. This will lead to different video frames for each video. For example, for this video (v_WallPushups_g21_c06), I may have 71 images, but you may have less than 71 images. So that's why your code could not load the data. To solve the problem, you need to generate your own train_rgb_split1.txt and val_rgb_split1.txt. You can write a simple python script to finish this task, basically just counting how many frames you have.

from hidden-two-stream.

Hi, Thank you so much for the help. I got almost less than 60 frames per video. if change number of frames in train_rgb_split1.txt and val_rgb_split1.txt will it effect the ground truth class? do i need change anything in testing datasets (UCF101 split 1, UCF101 split 2, UCF103 split 3) since i got different frames numbers for videos.

from hidden-two-stream.

It will effect the performance a little bit, but won't be much. You don't need to do anything, just update the train_rgb_split1.txt and val_rgb_split1.txt according to your situation.

from hidden-two-stream.

Hi, thanks for your answers. It helped me a lot.

from hidden-two-stream.

hello，
when i use the commond you gave ,When I was training, I made the following mistakes：
I0419 14:34:11.274951 2151 sgd_solver.cpp:106] Iteration 480, lr = 1e-06
I0419 14:36:48.555851 2151 solver.cpp:456] Snapshotting to binary proto file ./logs_end/vgg16_end_iter_500.caffemodel
I0419 14:37:00.028300 2151 sgd_solver.cpp:273] Snapshotting solver state to binary proto file ./logs_end/vgg16_end_iter_500.solverstate
I0419 14:37:04.153931 2151 solver.cpp:338] Iteration 500, Testing net (#0)
F0419 14:37:04.657557 2151 syncedmem.cpp:56] Check failed: error == cudaSuccess (2 vs. 0) out of memory
*** Check failure stack trace: ***
@ 0x7f7080db55cd google::LogMessage::Fail()
@ 0x7f7080db7433 google::LogMessage::SendToLog()
@ 0x7f7080db515b google::LogMessage::Flush()
@ 0x7f7080db7e1e google::LogMessageFatal::~LogMessageFatal()
@ 0x7f7081687d20 caffe::SyncedMemory::to_gpu()
@ 0x7f7081686d49 caffe::SyncedMemory::mutable_gpu_data()
@ 0x7f70816a3972 caffe::Blob<>::mutable_gpu_data()
@ 0x7f70816f7bb9 caffe::ScaleLayer<>::Forward_gpu()
@ 0x7f70816b3ec2 caffe::Net<>::ForwardFromTo()
@ 0x7f70816b3fd7 caffe::Net<>::Forward()
@ 0x7f708168e6d2 caffe::Solver<>::Test()
@ 0x7f708168f09e caffe::Solver<>::TestAll()
@ 0x7f708168f1c2 caffe::Solver<>::Step()
@ 0x7f708168fce9 caffe::Solver<>::Solve()
@ 0x40bb17 train()
@ 0x407724 main
@ 0x7f707f869830 __libc_start_main
@ 0x407ed9 _start
@ (nil) (unknown)
[6] 2151 abort (core dumped) ../../build/tools/caffe train -solver=end_solver.prototxt
the dataset is ucf and my cuda memory is about 8.1G

from hidden-two-stream.