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,
Steps to use this code:
Go to utils/ and run:
this downloads the darknet weight files. Also, fuses batchnorm layers and creates TensorFlow Ckpt files.
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.
To run Webcam inference:
Please refer the paper mentioned above to know more about the system used for testing and the versions of software tools used.