Quantization

Light version of convolutional neural network YoloV3 & DroNet for objects detection with a minimum of dependencies (INT8-inference)

Inspired and based on

• https://github.com/AlexeyAB/yolo2_light
• https://github.com/gplast/DroNet

Software Requirements

A Linux (Ubuntu 18.04) build environment needs these components:

• OpenCV 3.4 or higher
• GNU Compiler Collection (GCC) 3.4 or higher
• CMake 2.8 or higher
• Python 3.6 or higher

Clone Repository

cd ~
https://github.com/constandinos/quantization.git

Install

cd quantization
make
cd ..

How to compile:

• To compile for CPU just do make on Linux or build yolo_cpu.sln on Windows

• To compile for GPU set flag GPU=1 in the Makefile on Linux or build yolo_gpu.sln on Windows

  Required both CUDA >= 8.0 and cuDNN >= 7.1.1

Run (INT8-inference)

cd bin

# DroNet
./darknet detector test names/car.names cfg/DroNet_car.cfg weights/DroNet_car.weights images/car.png -thresh 0.4 -quantized

# DroNetV3
./darknet detector test names/car_ped.names cfg/DroNetV3_car.cfg weights/DroNetV3_car.weights images/car.png -thresh 0.4 -quantized

# Tiny-Yolo
./darknet detector test names/coco.names cfg/yolov3-tiny-car.cfg weights/yolov3-tiny-car.weights images/car.png -thresh 0.4 -quantized