Project 4c - YOLOv4 with CUDA support

Project start: 30 December 2021

Project finish: 1 January 2022

Objective: Installation of YOLO4 with CUDA support in VEnv

Yolo is a definition of a convolutional neural network which I believe has 106 layers.

Weights are calculated, i.e., the Yolo neural network is trained, using a program called Darknet. It is not a Python module but is a command line operation. Results are saved in a .weights file.

A .cfg file and .weights file are required to make use of the Yolo network. A .names file is also required. These will be explained later.

OpenCV reads these files for detection of objects.

Software installation:

add to .bashrc (located in HOME directory):

"/usr/local/cuda-10.2/bin:$PATH" (added to existing export path)

export LD_LIBRARY_PATH="/usr/local/cuda-10.2/lib64:$LD_LIBRARY_PATH"

close and open new terminal and check that nvcc is working:

nvcc -V

outside VEnv:

sudo apt-get update

sudo apt-get upgrade

clone the latest Darknet code for YOLO4 with CUDA support from GitHub:

(ref: https://jkjung-avt.github.io/yolov4/)

cd $HOME

cd Public/Projects/

change to the VEnv directory but do not activate it:

cd Public/Projects/myPython36VEnv

$ git clone https://github.com/AlexeyAB/darknet.git

$ cd darknet

vim Makefile:

Set the values: GPU=1 CUDNN=1

CUDNN_HALF=1 OPENCV=1

ARCH= -gencode arch=compute_53,code=[sm_53,compute_53]

make clean

make -j4  (takes about 1 minute)

If you have any problems or need clarification please contact me: jscanova@gmail.com