Installation

Clone the repository

git clone https://github.com/scosijnilvo/yolov8
cd yolov8

Create and activate a virtual env

python -m venv /path/to/venv
source /path/to/venv/bin/activate

Install with pip

pip install .

(or in editable mode if you want to modify the code)

pip install -e .

Dataset

Create dataset.yaml (edit paths, classes, and num_vars to fit your dataset)

path: ../yolov8/ # root dir
train: dataset/images/train # train images
val: dataset/images/val # val images
test: dataset/images/test # test images (optional)

# Classes
names:
  0: class_0
  1: class_1
  # ...

# Number of variables to predict
num_vars: 1

Each image must have a corresponding label file with the same name and .txt extension located at dataset/labels/[train|val|test]. The label files consist of one row for each object in the image with the following format.

For detection:

<class-index> <var_1> ... <var_n> <x> <y> <w> <h>

For segmentation:

<class-index> <var_1> ... <var_n> <x1> <y1> ... <xn> <yn>

where

• <class-index> = index of the class declared in the .yaml file
• <var_1> ... <var_n> = ground-truth values of the variables, set num_vars in the .yaml file to n
• <x> <y> <w> <h> = bounding box coordinates in xywh-format, normalized between 0 and 1
• <x1> <y1> ... <xn> <yn> = bounding coordinates of the segmentation mask, normalized between 0 and 1

Example code

# import
from ultralytics import RegressionModel

# training a detection model
model = RegressionModel('yolov8s-det-regression.yaml')
results = model.train(data='dataset.yaml', epochs=100)

# training a segmentation model
model = RegressionModel('yolov8s-seg-regression.yaml')
results = model.train(data='dataset.yaml', epochs=100)

# loading + evaluating on test set
model = RegressionModel('saved_model.pt')
metrics = model.val(split='test')

CLI

Training

Detection

yolo train model=yolov8s-det-regression.yaml data=dataset.yaml epochs=100

Segmentation

yolo train model=yolov8s-seg-regression.yaml data=dataset.yaml epochs=100

Validation

yolo val model=saved_model.pt data=dataset.yaml

Inference

yolo predict model=saved_model.pt source=image.jpg