search

thtrieu / darkflow

Translate darknet to tensorflow. Load trained weights, retrain/fine-tune using tensorflow, export constant graph def to mobile devices
GNU General Public License v3.0
6.13k stars 2.08k forks source link

Low FPS with Tesla K80 Graphics card and Xeon Server #1006

Open matthewvella90 opened 5 years ago

matthewvella90 commented 5 years ago

I'm processing a video in darkflow using yolo.cfg and yolov2.weights, and i'm getting very low fps, averaging 1.5fps. I'm running darkflow on a 56Gb ram Xeon server with a Tesla K80 graphics card, so setup is surely not an issue.

I'm running using the command:

python flow --model cfg/yolo.cfg --load bin/yolov2.weights --demo vid.mp4 --gpu 1.0 --saveVideo. Console output:

GPU mode with 1.0 usage 2019-03-21 12:59:26.722897: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 Finished in 13.213001251220703s

Press [ESC] to quit demo 1.497 FPS End of Video

Demo stopped, exit.

What is the problem for such low FPS?

skofot commented 5 years ago

Can you verify that the GPU is actually being used? 1.5 FPS is pretty much the same as I get when using CPU. It might be that CUDA is not installed, which is required for TensorFlow GPU. I recommend reading https://www.tensorflow.org/install/gpu