Giters

alew3 / faceit_live3

This is an update to faceit_live using first order model

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Could not find a version that satisfies the requirement

Migushthe2nd opened this issue · comments

commented

I have tried Python 3.6, 3.7 and 3.8, but for all three a version that satisfies the requirement could not be found for the following modules:

  • mkl-fft==1.0.15
  • mkl-random==1.1.0
  • mkl-service==2.3.0
  • pandas==1.0.3
  • Pillow==7.0.0
  • prometheus-client==0.7.1
  • prompt-toolkit==3.0.5
  • torch==1.4.0
    And at last
  • astroid 2.3.3 has requirement wrapt==1.11.*, but you'll have wrapt 1.12.1 which is incompatible.
commented

I'm not sure this is Conda related, but I will, just to be sure, try this tomorrow.

commented

What OS and Anaconda version are you running?

commented

I'm on Windows 10, but I apparently also had miniconda installed, which was configured in my Path variable. Also, I now ran the commands via the anaconda prompt, and all requirement installed fine.

Things I noticed:

  • The python script prints a message which says R can be used to switch between relative and absolute, while it should say T
  • There is no clear indication in the command prompt a key was pressed (N, C, T, Q)
  • CPU usage is extremely high. I'm using an Intel I5-9600K (hexa-core) at 5.0GHz and it's usage is 100%, while gpu usage is relatively low (only 6-8% of my RTX 2060). There's quite a lot performance headroom on the gpu.
  • Changing python affinity to only use core 6 decreases cpu usage by 60% with no noticeable change in quality.
commented

Thanks for pointing it out about the text, I changed it on the README, but had forgotten to change it in the code message.

I'm not much of a Windows user, so I did a quick and dirty test yesterday to see if it would run with some changes. It did, but it is much slower than the Linux version and I couldn't get it to fully utilize my GPU also. Not sure if this is a Pytorch thing on Windows or a CPU bottleneck because of the preprocessing.

commented

@Migushthe2nd apparently Task Manager doesn't report correctly the GPU usage on Windows. You need to use nvidia-smi.exe to view the correct GPU usage % (you can find it at C:\Program Files\NVIDIA Corporation\NVSMI)

On my Quad-Core 6700K + Titan RTX it was using 26%, so there is definately a CPU bottleneck on my setup.

commented

Hmm yeah it indeed uses more gpu. Anyways, thank you for your quick assistance!

Also, just noticed another typo at line 120 in faceit_live.py:
cv2.resizeWindow('Stream', webcam_width,webcam_width)
should be
cv2.resizeWindow('Stream', webcam_width,webcam_height)
See my pr

commented

What OS and Anaconda version are you running?

please can you make a youtube video on the tutorials on how to do it please?