Skip to content

Latest commit

 

History

History
35 lines (23 loc) · 1.19 KB

README.md

File metadata and controls

35 lines (23 loc) · 1.19 KB

YOLO-Object-Detection optimization on Xeon scalable processors

In this project, optimization of TensorFlow code is performed for an object detection application to obtain real-time performance.
Please refer the following paper for all the details regarding performance optimizations,
https://colfaxresearch.com/yolo-optimization/

Rquirements:

Numpy
Python 2.7
Tensroflow
OpenCV

Steps to use this code:

  1. Go to utils/ and run:
    $ python config.py
    this downloads the darknet weight files. Also, fuses batchnorm layers and creates TensorFlow Ckpt files.

  2. To run image inference:
    $ python inference.py , to run TinyYolo model
    $ python inference.py --image= [image path]
    $ python infernce.py --v2 , to run YoloV2 model
    $ NUM_INTER_THREADS=2 NUM_INTRA_THREADS=8 python inference.py --par , to run parallel TensorFlow session(Inter/Intra op threads), if it is supported in your system.

  3. To run Webcam inference:
    $ python webcam_inference.py

Please refer the paper mentioned above to know more about the system used for testing and the versions of software tools used.