error for video format when training kinetics from scratch

[InwoongLee]

I'm trying to train kinetics from scratch. I finished caffe2 from source with USE_OPENCV=1 USE_FFMPEG=1 USE_LMDB=1. However, I received these errors.

[E video_decoder.cc:150] Insufficient data to determine video format
[E video_decoder.cc:751] The video seems faulty and we could not decode enough frames: 0 VS 8
INFO:train_net:Finished iteration 0/1000000 of epoch 0 (0.65 clips/sec)
INFO:train_net:Training loss: 5.99146461487, lr: 0.00999999977648, accuracy: 0.0
[E video_decoder.h:269] Error opening video file ^@^@^@^\ftypisom^@^@^B^@isomiso2mp41^@^@^@^Hfree^@?3�mdat�^D^@Lavc57.107.100^@bv^_���8^@H��^X:^Z^^II�`�X:GK��ܷ^Nt��޺Q�2�K�/n���gpw���W^^��^Y"^NIU�^S��T�hJ^H�HGN�^E1^Dt;^N@����M�y��G^Vs^D��
�^_�fRc^_��=6��L'�z�/^Y��^G�� �La^H��Wn���2I�B_G�^\�^X,g?>ܶ
^Q^N^Ln�s�9G6xL^Z^Rl�^Q6l#^@^pqf^?

I don't know how to error for opening video file.
Anybody knows this?

[dutran]

This is either your videos has no frame or the parameter setting is incorrect. Please pay attention to use_list (in create_video_db.py) and use_local_file (in train_net.py). These parameters are needed to be used together, meaning both be 0 or 1.

[InwoongLee]

I set use_list to 0 in create_video_db.py and use_local_file in train_net.py.
Actually, that is the default settings. Can you guess this problem's reason?
Thank you for your reply.

[dutran]

From you error message, it looks like you use use_local_file=1 that's why it said a weird filename in the log. To debug, you may want to try again with both use_list=1 and use_local_file=1