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.