Saltar a contenido

Installation

Introduction

Introduction

This module demonstrates how to detect objects in images using the deep learning network You Only Look Once version 8 (YOLOv8).

The computer vision content is divided into five lessons:

  1. Installation: In this lesson, you will install all the necessary tools and libraries to work with computer vision in MATLAB.
  2. Training Models using Transfer Learning: This lesson covers the basics of training a YOLO model and explains how transfer learning can improve performance and reduce training time.
  3. Post-processing: This lesson presents post-processing techniques that enhance model performance after training, including filtering and refinement of detections.
  4. Evaluation: This lesson focuses on the evaluation of object detection models using standard metrics and visualization techniques. It covers the computation of these metrics and provides guidelines for their proper interpretation.
  5. Experimentation: This final lesson provides a deeper exploration of model training. It introduces key hyperparameters, demonstrates the application of data augmentation techniques, and explains how to design experiments to evaluate different training configurations.

Required Toolboxes

To run this project or lesson, the following toolboxes are required:

  • Computer Vision Toolbox™
  • Deep Learning Toolbox™
  • Parallel Computing Toolbox™

Check Toolbox Installation in MATLAB

You can use the code below to automatically check if all required toolboxes are installed:

% List of required toolboxes
requiredToolboxes = [
    "Computer Vision Toolbox"
    "Deep Learning Toolbox"
    "Parallel Computing Toolbox"
];

% Get installed toolboxes and convert names to string array
toolboxTable = matlab.addons.installedAddons;
installedNames = string(toolboxTable.Name);

% Check for missing toolboxes
missing = ~ismember(requiredToolboxes, installedNames);

% Report
if any(missing)
    missingList = requiredToolboxes(missing);
    error("Missing required toolboxes:\n%s\n\nPlease install them via Home > Add-Ons > Get Add-Ons.", ...
        strjoin(missingList, newline));
else
    disp("✅ All required toolboxes are installed.");
end

✅ All required toolboxes are installed.

How to Install Missing Toolboxes

If any toolbox is missing:

  1. Go to the Home tab in MATLAB.
  2. Click on Add-Ons > Get Add-Ons.
  3. Search for the missing toolbox by name.
  4. Click Install.

Setup Instructions

Add Help Module Path

Add the path to the help module so that MATLAB can access the scripts in this folder:

addpath('help-module');

Install Python and Required Libraries

Use the following command to install Python and the necessary libraries. This may take more than 5 minutes:

yolo.installUltralytics();

After execute this code, it should appear a win64 in case you are using windows, else glnxa64.

Download the Dataset

To download the dataset:

(It is not updated, is other dataset, don't download this)

%disp("download data")

download data

%yolo.helper.downloadMultiClassData()

Downloading multiclass object detection dataset...

Testing YOLO Is Working

Load the YOLOv8 Object Detector and Run Inference:

% Load YOLO v8 model
det = yolov8ObjectDetector2('yolov8s');

Pretrained yolov8s network already exists.

% Read test image.
I = imread(fullfile('datasets/fruits_3_4998/example','orange.jpg'));

% Load YOLO v8 small network.
det = yolov8ObjectDetector2('yolov8s');

Pretrained yolov8s network already exists.

% Perform detection using pretrained model.
[bboxes, scores, labels] = detect(det, I);

% Visualize detection results.
annotations = string(labels) + ': ' + string(scores);
Iout = insertObjectAnnotation(I, 'rectangle', bboxes, annotations);
figure, imshow(Iout);

figure_0.png

YOLO Settings

Set Python Environment

if (ispc)
    pyenv(Version="win64/python/python.exe", ExecutionMode = "OutOfProcess")
else
    pyenv(Version="glnxa64/python/bin/python3", ExecutionMode = "OutOfProcess")
end

ans = 
  PythonEnvironment with properties:

          Version: "3.11"
       Executable: "C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\from-code-to-robot\computer_vision\win64\python\python.exe"
          Library: "C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\from-code-to-robot\computer_vision\win64\python\python311.dll"
             Home: "C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\from-code-to-robot\computer_vision\win64\python"
           Status: Loaded
    ExecutionMode: OutOfProcess
        ProcessID: "30028"
      ProcessName: "MATLABPyHost"

if isunix
    py.sys.setdlopenflags(int32(bitor(int64(py.os.RTLD_LAZY),int64(py.os.RTLD_DEEPBIND))));
end

Configure YOLO Global Settings

% setYoloSettings Set YOLO global settings via Python wrapper.
%
% setYoloSettings(pythonObject, option, value) sets the specified YOLO
% option to the given value using the Python wrapper.
py.trainYOLOv8Wrapper.setYoloSettings(py.str('tensorboard'), true);

settings_version 0.0.4
datasets_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\Lessons\from-code-to-robot\computer_vision\datasets
weights_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\Lessons\from-code-to-robot\computer_vision\weights
runs_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\Lessons\from-code-to-robot\computer_vision\runs
uuid ec83097be249f049dc01533308f726e37a20a9a11e43afb8810c32587f8ec051
sync True
api_key 
openai_api_key 
clearml True
comet True
dvc True
hub True
mlflow True
neptune True
raytune True
tensorboard True
wandb True
True

runsDir = fullfile(pwd, 'runs'); % Une el directorio actual con la carpeta 'runs'
weightsDir = fullfile(pwd, 'weights'); % Une el directorio actual con la carpeta 'runs'
datasets_dir = fullfile(pwd, 'datasets');
py.trainYOLOv8Wrapper.setYoloSettings(py.str("runs_dir"), py.str(runsDir));

settings_version 0.0.4
datasets_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\Lessons\from-code-to-robot\computer_vision\datasets
weights_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\Lessons\from-code-to-robot\computer_vision\weights
runs_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\Lessons\from-code-to-robot\computer_vision\runs
uuid ec83097be249f049dc01533308f726e37a20a9a11e43afb8810c32587f8ec051
sync True
api_key 
openai_api_key 
clearml True
comet True
dvc True
hub True
mlflow True
neptune True
raytune True
tensorboard True
wandb True
True

py.trainYOLOv8Wrapper.setYoloSettings(py.str("weights_dir"), py.str(weightsDir));

settings_version 0.0.4
datasets_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\Lessons\from-code-to-robot\computer_vision\datasets
weights_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\Lessons\from-code-to-robot\computer_vision\weights
runs_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\from-code-to-robot\computer_vision\runs
uuid ec83097be249f049dc01533308f726e37a20a9a11e43afb8810c32587f8ec051
sync True
api_key 
openai_api_key 
clearml True
comet True
dvc True
hub True
mlflow True
neptune True
raytune True
tensorboard True
wandb True
True

py.trainYOLOv8Wrapper.setYoloSettings(py.str("datasets_dir"), py.str(datasets_dir));

settings_version 0.0.4
datasets_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\Lessons\from-code-to-robot\computer_vision\datasets
weights_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\from-code-to-robot\computer_vision\weights
runs_dir C:\Users\Noel Nathan\Desktop\Universidad\8tQuadrimestre\tfg\from-code-to-robot\computer_vision\runs
uuid ec83097be249f049dc01533308f726e37a20a9a11e43afb8810c32587f8ec051
sync True
api_key 
openai_api_key 
clearml True
comet True
dvc True
hub True
mlflow True
neptune True
raytune True
tensorboard True
wandb True
True