Strange error when extracting rgb-images and optical flow by dense flow.

hi @zhaoyue-zephyrus

when using dense_flow to exact frames and optical flow for kinetics datasets,most of video can be successfully extracted，but others maybe will report an same error as follows resulting in extraction failure:

terminate called after throwing an instance of ‘cv::Exception’ what(): OpenCV(4.1.0) /home/lw/opencv-4.1.0/modules/core/src/cuda_info.cpp:73: error: (-217:Gpu API call) invalid device ordinal in function ‘setDevice’ Aborted (core dumped)

to these semmingly problematic video , when I using the following commands(suppose the video was renamed to test.mp4) to extract rbg-frames and optical flow individually:

./extract_gpu_ -f=test.mp4 -x=tmp/flow_x -y=tmp/flow_y -i=tmp/image -b=20 -t=1 -d=0 -s=1 -o=dir

it will be no problem!!

my dataset’s directory as follows： |------kinetics | |--------------abseiling | | |---------------------_4YTwq0-73Y_000044_000054.mp4 | | |---------------------3B32lodo2M_000059_000069.mp4 | | |-----------------------… | |--------------air_drumming | | |-------------------.03V2idM7_KY_000003_000013.mp4 | | |--------------------… | |--------------… | |--------------zumba | | |----------------- klz-HIvmlJo_000025_000035.mp4 | | |-----------------… | | |------------------ klz-HIvmlJo_000025_000035.mp4

[zhaoyue-zephyrus]

@longweiwei

1. The first problem arises because num_gpu is greater than the number of gpu you actually have on your machine.

2. I am not quite clear about your saying run multiple programes using same executable file? Can you explain more?

[zhaoyue-zephyrus]

@longweiwei I don’t think that the practice you mentioned may cause any problem. Actually the multiprocessing in the script calls extract_gpu in parallel. To modify the number of frames, consider changing the argument -t=xx which is 1 by default.