Autodistill YOLOE

Description

Use YOLOE to auto-label images for use in training fine-tuned object detection models

Overview

Autodistill YOLOE is a package that allows you to use YOLOE models with the Autodistill framework. YOLOE is a state-of-the-art object detection model that supports both text/visual prompting and prompt-free detection.

This integration enables you to:

1. Use YOLOE models to automatically label images
2. Prepare datasets for training custom object detection models
3. Leverage both text-prompted and prompt-free detection capabilities

Installation

Python 3.11 or later is required

pip install autodistill-yoloe

Usage

Basic Example

from autodistill_yoloe import YOLOEBase
from autodistill.detection import CaptionOntology

# Define your ontology (the objects you want to detect)
ontology = CaptionOntology({"person": "person", "car": "car", "dog": "dog"})

# Initialize the YOLOE model
yoloe = YOLOEBase(ontology=ontology)

# Label a folder of images
dataset = yoloe.label("path/to/images", extension=".jpg")

# The labeled dataset is ready for training a custom model
print("Dataset created and ready for distillation!")

Using a Custom YOLOE Model

from autodistill_yoloe import YOLOEBase
from autodistill.detection import CaptionOntology

ontology = CaptionOntology({"person": "person", "car": "car"})

# Use a specific YOLOE model
yoloe = YOLOEBase(ontology=ontology, model="yoloe-11s-seg.pt")

# Make predictions on a single image
detections = yoloe.predict("path/to/image.jpg")
print(detections)

Human-in-the-Loop Labeling with Roboflow

from autodistill_yoloe import YOLOEBase
from autodistill.detection import CaptionOntology

ontology = CaptionOntology({"person": "person", "car": "car"})
yoloe = YOLOEBase(ontology=ontology)

# Enable human-in-the-loop labeling with Roboflow
dataset = yoloe.label(
    "path/to/images",
    human_in_the_loop=True,
    roboflow_project="my-project-name"
)

Available Models

YOLOE supports several model variants:

Text/Visual Prompt Models

• YOLOE-11S
• YOLOE-11M
• YOLOE-11L
• YOLOE-v8S
• YOLOE-v8M
• YOLOE-v8L

Prompt-Free Models

• YOLOE-11S-PF
• YOLOE-11M-PF
• YOLOE-11L-PF
• YOLOE-v8S-PF
• YOLOE-v8M-PF
• YOLOE-v8L-PF

Advanced Features

• SAHI Integration: Enable SAHI (Slicing Aided Hyper Inference) for improved detection of small objects
• NMS Settings: Configure Non-Maximum Suppression for better detection results
• Confidence Recording: Save confidence scores with detections

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgements

• Autodistill - The framework that makes this integration possible
• Ultralytics - Creators of the YOLOE model
• Supervision - Computer vision toolkit used for annotations