Use face_recognition on raspberry, “segmentation fault” occurred

[sxyzzzzzzz]

• face_recognition version:0.2.0
• Python version:3.5
• Operating System:raspbian

python facerec_on_raspberry_pi.py

segmentation fault

[sxyzzzzzzz]

Similar to https://github.com/ageitgey/face_recognition/issues/119

[sxyzzzzzzz]

cpu 100%，then segmentation fault. raspberry pi 3.

[sxyzzzzzzz]

seems to freeze the function "face_encodings"

[jparedesj]

Do you found a solution for this?

[andrewda]

Any solution found for this? Been debugging for many hours now to no avail.

[xyG67]

Have you found the solution? I met the same problem

[andrewda]

Unfortunately I never found a solution to this, besides using some faster hardware. I'd love to hear if others have figured this out, though!

[andrewplesniak]

I was experiencing the same problem, segmentation fault when calling the face_encodings on the raspberry pi 3. In my case, I compiled dlib with OpenBLAS for linear algebra optimizations which allows dlib functionality to execute faster. However, OpenBLAS itself is multi-threaded and if you try and use it in a muilt-threaded application it will peg the CPU and cause the seg fault. For use in multi-threaded applications, OpenBlas recommends you set it to single thread mode.

In my case, I set some environmental variables in my virtual environment to tell OpenBLAS to only use a single thread by doing the following:

$ workon myvenv $ nano VIRTUAL_ENV/bin/postactivate

add the following lines and save: export OPENBLAS_NUM_THREADS=1 export OPENBLAS_MAIN_FREE=1

By doing this, I was able to achieve ~15 fps of facial recognition in my application that was previously crashing on my pi.

[AlexBdx]

I was experiencing the same problem, segmentation fault when calling the face_encodings on the raspberry pi 3. In my case, I compiled dlib with OpenBLAS for linear algebra optimizations which allows dlib functionality to execute faster. However, OpenBLAS itself is multi-threaded and if you try and use it in a muilt-threaded application it will peg the CPU and cause the seg fault. For use in multi-threaded applications, OpenBlas recommends you set it to single thread mode.

In my case, I set some environmental variables in my virtual environment to tell OpenBLAS to only use a single thread by doing the following:

$ workon myvenv $ nano VIRTUAL_ENV/bin/postactivate

add the following lines and save: export OPENBLAS_NUM_THREADS=1 export OPENBLAS_MAIN_FREE=1

By doing this, I was able to achieve ~15 fps of facial recognition in my application that was previously crashing on my pi.

This solved my issue too. The Pi is now using only one CPU core instead of 4 but does not crash when calling face_encodings! Any idea how all 4 cores could be used to increase the FPS?

[SeanIsAllRobot]

I have the same issue, but I'm not working within a virtual environment, so where would i write

export OPENBLAS_NUM_THREADS=1
export OPENBLAS_MAIN_FREE=1

to?