dronefreak/human-action-classification

Inference speed

lisa676 opened this issue · 1 comments

lisa676 commented

Hi @dronefreak Thanks for this wonderful repository. I just run it in my PC it's working fine and evaluation time for single image is about 0.2-0.3s which is really fast but when I run webcam then output is quite slow I don't know what is the problem. My PC configuration is

  • Ubuntu 16

  • CUDA 9

  • Python 3.7 with TF 1.15.1 and OpenCV 4.1

I have suggestion that also add a piece of code for video from personal directory

dronefreak commented

Hi,

Apologies for the late reply. The code basically has to instantiate the GPU twice for every pose computation and scene classification, as I am yet to combine them into a single tf.Session(). In that case the instantiation would take place only once for both.
The method suggested in this repo was part of an academic project, and I have not had the time to debug the instantiation issue. Please let me know if this helps with your query.