Installation of CUDA and its requirement to run YOLOv3-v4

YOLO: You only look once (YOLO) is a state-of-the-art, real-time object detection system.

Get detections on a single pass.

1. Resize the input image (e.g., 448x448).
2. Run Convolutional Network to get detections.
3. Filter the output by a Non-max supression algorithm (to remove multiple detections of the same object)

Architecture (training phase):

• Input size: 448 x 448
• Nine Convolutional layers with leaky ReLU.
• Six max pooling layers.
• Backbone feature map: 7 x 7 (7 x 64 = 448)

Output size: 30 x 7 x 7:

• Two bounding box definitions, containing x, y, height, width andconfidence. (2 x 5)
• Class probabilities, relevant if only object confidence is high enough. (20)

Box probabilities:

If there is an object in a cell, we decide which object by finding the biggest probability.

Non-Max Supreesion:

• An object may be detected several times by different boxes (grid cells).
• So, we need to use non-max supression to filter the results.
• Final detections are obtained by using thresholding and non-max supression.

For more information see the Darknet: Open Source Neural Networks in C

source code: https://github.com/AlexeyAB/darknet

Here, I guide you step by step with a bare machine to get a real time object detector with YOLO v3 and V4.

Installation Summary:

• Install Ubuntu 18.03
• Install Git
• Install Cmake >= 3.12
• Install CUDA 10.2 (Nvidia GPU only!)
• Install cuDNN 7.6.4 (compatible for CUDA 10.1)
• Install Python3 and Pip3
• Install OpenCV 3.2.0
• Install GC

A Ubuntu 18.04 native system is preferred in training process. At least one NVIDIA GPU Card is required such as GeForce series to enable GPU mode. One can train a network with custom data but for a videa you will end up killing the process as it may take days.

I recently tested running YOLo on both 12 CPU and GPU. Here are the steps you may follow:

(a) Environment Create a virtual environment and start installing dependencies and libraries.

(b) Requirements

# Install Git
apt-get install git
git --version

# Install Cmake >= 3.12
apt install -y cmake
cmake --version

Install CUDA NOTE: YOLO Darknet from AlexeyAB repository can be used with CUDA 10.2. To install cuda you can also visit CUDA Toolkit. On terminal shell you can follow these steps:

sudo wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin

sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600

sudo wget http://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda-repo-ubuntu1804-10-2-local-10.2.89-440.33.01_1.0-1_amd64.deb

sudo dpkg -i cuda-repo-ubuntu1804-10-2-local-10.2.89-440.33.01_1.0-1_amd64.deb

sudo apt-key add /var/cuda-repo-10-2-local-10.2.89-440.33.01/7fa2af80.pub

sudo apt-get update

sudo apt-get -y install cuda

#NOTE: Restart Ubuntu after CUDA installed!!

Then set-up CUDA 10.2 path

echo "export PATH=/usr/local/cuda-10.2/bin${PATH:+:${PATH}}" >> ~/.bashrc
source ~/.bashrc

To verify CUDA is installed successfully, run the comand ncc --version, and `nvidia-sim.

Install cuDNN

Next, install cuDNN 7.6.4 (compatible for CUDA 10.1) see also. You can download cuDNN Runtime Library for Ubuntu18.04 (Deb), cuDNN Developer Library for Ubuntu18.04 (Deb), cuDNN Code Samples and User Guide for Ubuntu18.04 (Deb) also these three libraries manually and save in the same folder and run the command:

sudo dpkg -i libcudnn7_7.6.4.38-1+cuda10.1_amd64.deb
sudo dpkg -i libcudnn7-dev_7.6.4.38-1+cuda10.1_amd64.deb
sudo dpkg -i libcudnn7-doc_7.6.4.38-1+cuda10.1_amd64.deb

To verify that cuDNN is installed and is running properly, compile the mnistCUDNN sample located in the /usr/src/cudnn_samples_v7 directory in the Debian file see.

Procedure

cp -r /usr/src/cudnn_samples_v7/ $HOME
cd $HOME/cudnn_samples_v7/mnistCUDNN
make clean && make
./mnistCUDNN

If cuDNN is properly installed and running on your Linux system, you will see a message similar to the following: Test passed!

Next, we can upgrade pip and install OpenCV,

sudo python -m pip install --upgrade pip
sudo pip3 install opencv-python

# Verify installation
# python (enter python shell)
# >> import cv2
# >> print (cv2.__version__)

Also, one can check the gcc version with command gcc ---version.

For YOLO V3, I follow the steps from here, YOLO V3 and run a test image as shown here.

# Install Git 
# sudo apt-get install git
git clone https://github.com/pjreddie/darknet
cd darknet
make

#download the pre-trained weight file
wget https://pjreddie.com/media/files/yolov3.weights 

#Then run the detector!
./darknet detect cfg/yolov3.cfg yolov3.weights data/dog.jpg

#Easy!Done, You can make a your own image 

Here is repositories for YOLO V4 darknet.

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

Download yolov4.weights file 245 MB (https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v3_optimal/yolov4.weights).

Before, you compile, make few changes in Makefile if you wish to run on GPU GPU=1, CUDNN=1, CUDNN_HALF=1. Otherwise you can unable it. (see make files). Then run the command make clean and then make. Note everytime, you do any changes you first run make clean. Now the executable file 'darknet' is created nd run the darknet:

./darknet detector test cfg/coco.data cfg/yolov4.cfg yolov4.weights data/dog.jpg

to have a video you must use GPU:

./darknet detector demo cfg/coco.data cfg/yolov4.cfg yolov4.weights data/***.mp4

Here is the output of my image data/4.JPG: Predicted in 1711.256000 milli-seconds. zebra: 100% person: 99% handbag: 60% person: 28%

Yolo-V4 is faster and nicer, is not it?

Original Code: https://github.com/pjreddie/darknet

Site: https://pjreddie.com/darknet/yolo/