OpenCV 4.2.0 with python3 and CUDA : Segfault on cv2.VideoCapture()

0
votes

I encounter a problem with OpenCV that I have for several days now : it segfaults when calling the cv2.VideoCapture() function.

When launching my script (with GDB) :

extract-all_1   | Thread 1 "python3" received signal SIGSEGV, Segmentation fault.
extract-all_1   | 0x00007f83857fe33b in bool pyopencv_to<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >(_object*, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >&, ArgInfo const&) [clone .isra.1286] ()
extract-all_1   |    from /usr/lib/python3/dist-packages/cv2/python-3.6/cv2.cpython-36m-x86_64-linux-gnu.so
extract-all_1   | (gdb) quit

When running my script without GDB, the container exits with code 139

I identified the problem occures when calling the "cv2.VideoCapture()" function :

def perform_video_extraction(video_path):
    input_movie = cv2.VideoCapture(video_path)
    nb_total_frames = int(input_movie.get(cv2.CAP_PROP_FRAME_COUNT))
    [...]

Hints :

  • I process MP4 video files
  • I've tried compressing my videos that are >30fps to 25fps
  • I've tried with OpenCV 3.4.9, 4.1.0, 4.1.1, 4.1.2, 4.2.0 and 4.3.0 (pip install)
  • I've tried compiling OpenCV 4.2.0 and 4.3.0 from source
  • I've tried each version above successively with CUDA 10.0, 10.1 and 10.2 : each version for each case produces the same error
  • This segfault is not reproduced when using the CPU (non-cuda) version of OpenCV

Here is my Dockerfile (CUDA 10.2 with OpenCV 4.2.0 built from source) : https://pastebin.com/raw/a42wtcRG

Here is what the cmake summary build returns : https://pastebin.com/raw/SFPUakyL

My config :

Have you any recommendation for debugging this problem ?

Thank you

1
dockeropencvpython-3.6nvidia-docker

1 Answers

1
votes

I managed to debug the problem. Due to a stupid encoding issue.

Adding :

ENV LANG C.UTF-8

to my Dockerfile managed to make the container run (my original pastebin mentioned this line but after doublecheck, I didn't have it).

I was able to find out this idea because of this more accurate backtrace from GDB : 

root@f42846d26d89:/opencv-4.2.0/build# gdb --args python3 -u /usr/app/scripts/extract.py
GNU gdb (Ubuntu 8.1-0ubuntu3.2) 8.1.0.20180409-git
Copyright (C) 2018 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-linux-gnu".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.
Find the GDB manual and other documentation resources online at:
<http://www.gnu.org/software/gdb/documentation/>.
For help, type "help".
Type "apropos word" to search for commands related to "word"...
Reading symbols from python3...(no debugging symbols found)...done.
(gdb) run
Starting program: /usr/bin/python3 -u /usr/app/scripts/extract.py
warning: Error disabling address space randomization: Operation not permitted
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
[...]
Thread 1 "python3" received signal SIGSEGV, Segmentation fault.
getUnicodeString (str="", obj=<optimized out>) at /opencv-4.2.0/modules/python/src2/pycompat.hpp:69
69          if (PyBytes_Check(bytes))
(gdb) backtrace
#0  0x00007f2959a1433b in getUnicodeString (str="", obj=<optimized out>) at /opencv-4.2.0/modules/python/src2/pycompat.hpp:69
#1  0x00007f2959a1433b in pyopencv_to<std::__cxx11::basic_string<char> >(PyObject*, cv::String&, ArgInfo const&) (obj=<optimized out>, value="", info=...)
    at /opencv-4.2.0/modules/python/src2/cv2.cpp:731
#2  0x00007f2959dd6a2d in pyopencv_cv_VideoCapture_VideoCapture(pyopencv_VideoCapture_t*, PyObject*, PyObject*) (self=0x7f2965344190, args=0x7f296307c3c8, kw=0x0)
    at /opencv-4.2.0/build/modules/python_bindings_generator/pyopencv_generated_types_content.h:21272
#3  0x0000000000551b81 in  ()
#4  0x00000000005aa6ec in _PyObject_FastCallKeywords ()
#5  0x000000000050abb3 in  ()
#6  0x000000000050c5b9 in _PyEval_EvalFrameDefault ()
#7  0x0000000000509d48 in  ()
#8  0x000000000050aa7d in  ()
#9  0x000000000050c5b9 in _PyEval_EvalFrameDefault ()
#10 0x0000000000508245 in  ()
#11 0x000000000050b403 in PyEval_EvalCode ()
#12 0x0000000000635222 in  ()
#13 0x00000000006352d7 in PyRun_FileExFlags ()
#14 0x0000000000638a8f in PyRun_SimpleFileExFlags ()
#15 0x0000000000639631 in Py_Main ()
#16 0x00000000004b0f40 in main ()
(gdb) list
64  {
65      bool res = false;
66      if (PyUnicode_Check(obj))
67      {
68          PyObject * bytes = PyUnicode_AsUTF8String(obj);
69          if (PyBytes_Check(bytes))
70          {
71              const char * raw = PyBytes_AsString(bytes);
72              if (raw)
73              {
(gdb)

/opencv-4.2.0 being my install path

It seems like my filenames were not in a right encoding format.

Finally, I specify that pip installing the python binding directly works perfectly fine now this modification has been brought.