YOLOv8 Human Labeling in MATLAB using Ground Truth Labeler

This guide explains how to integrate a pretrained YOLOv8 model with MATLAB's Ground Truth Labeler App to automatically generate human bounding box annotations and manually assign IDs using attributes.

---

Requirements

• MATLAB R2023b or later
• Toolboxes:
  • Computer Vision Toolbox
  • Deep Learning Toolbox
  • Deep Learning Toolbox Converter for ONNX Model Format

---

Steps to use YOLOv8 with Ground Truth Labeler in MATLAB

1. Download the repository

Download the YOLOv8 model integration from:

🔗 https://github.com/matlab-deep-learning/Pretrained-YOLOv8-Network-For-Object-Detection.git

---

2. Add the folder to MATLAB path

In MATLAB, navigate to the downloaded folder in the Current Folder pane and run:

addpath(genpath(pwd));

---

3. Install required toolboxes

Go to the Home tab → click Add-Ons → Get Add-Ons. In the Add-On Explorer, search for and install the following:

• Computer Vision Toolbox
• Deep Learning Toolbox
• Deep Learning Toolbox Converter for ONNX Model Format

---

4. Prepare your files

• Place the custom automation file yolov8_label.m in the same folder.
• Add your .png image folder (e.g., output_images) and the file load_timestamps.m (from this repo: https://github.com/Cyano0/Label-the-AoC-Rosbag-Data/tree/main) into the same directory.
• Run the following command in MATLAB:

load_timestamps;

---

5. Open the Ground Truth Labeler App

In the MATLAB Command Window, run:

groundTruthLabeler

---

6. Load your images

• Load the folder containing your .png frames (e.g., output_images).

---

7. Create a label

• Add a new rectangle ROI label named "human".

---

8. Import the custom algorithm

• Go to Select Algorithm → click Import Algorithm → choose yolov8_label.m.

---

9. Select the custom algorithm

• Choose "YOLOv8 Human Labeler" from the automation list.

---

10. Run the automation

• Click Automate → then click Run. YOLOv8 will automatically draw bounding boxes for detected humans.

---

11. Accept the results

• Once satisfied, click Accept to keep the generated labels.

---

12. Add an ID attribute

• Click the checkbox next to the label "human" in the label list.
• Click Attribute → Add a new attribute called ID.
• Set the Type to "List" and provide values like 1, 2, 3, ....

---

13. Assign IDs to detections

For each frame:

• Click on each bounding box.
• In the right panel, assign the correct ID (e.g., 1 for person1, 2 for person2, etc.).

---