autodistill / autodistill-yolov8

YOLOv8 Target Model plugin for Autodistill

Autodistill: YOLOv8 Target Model

This repository contains the code implementing YOLOv8 as a Target Model for use with autodistill. You can also use a YOLOv8 model as a base model to auto-label data.

YOLOv8 is a Convolutional Neural Network (CNN) that supports realtime object detection, instance segmentation, and other tasks. It can be deployed to a variety of edge devices.

Read the full Autodistill documentation.

Read the YOLOv8 Autodistill documentation.

Installation

To use the YOLOv8 Target Model, simply install it along with a Base Model supporting the detection task:

pip3 install autodistill-grounded-sam autodistill-yolov8

You can find a full list of detection Base Models on the main autodistill repo.

Quickstart (Train a YOLOv8 Model)

from autodistill_grounded_sam import GroundedSAM
from autodistill.detection import CaptionOntology
from autodistill_yolov8 import YOLOv8

# define an ontology to map class names to our GroundingDINO prompt
# the ontology dictionary has the format {caption: class}
# where caption is the prompt sent to the base model, and class is the label that will
# be saved for that caption in the generated annotations
base_model = GroundedSAM(ontology=CaptionOntology({"shipping container": "container"}))

# label all images in a folder called `context_images`
base_model.label(
  input_folder="./images",
  output_folder="./dataset"
)

target_model = YOLOv8("yolov8n.pt")
target_model.train("./dataset/data.yaml", epochs=200)

# run inference on the new model
pred = target_model.predict("./dataset/valid/your-image.jpg", confidence=0.5)
print(pred)

# optional: upload your model to Roboflow for deployment
from roboflow import Roboflow

rf = Roboflow(api_key="API_KEY")
project = rf.workspace().project("PROJECT_ID")
project.version(DATASET_VERSION).deploy(model_type="yolov8", model_path=f"./runs/detect/train/")

Quickstart (Use a YOLOv8 Model to Label Data)

from autodistill_yolov8 import YOLOv8Base
from autodistill.detection import CaptionOntology

# define an ontology to map class names to our YOLOv8 classes
# the ontology dictionary has the format {caption: class}
# where caption is the prompt sent to the base model, and class is the label that will
# be saved for that caption in the generated annotations
# then, load the model

# replace weights_path with the path to your YOLOv8 weights file
base_model = YOLOv8Base(ontology=CaptionOntology({"car": "car"}), weights_path="yolov5s.pt")

# run inference on a single image
results = base_model.predict("mercedes.jpeg")

base_model.label(
  input_folder="./images",
  output_folder="./dataset"
)

Choosing a Task

YOLOv8 supports training both object detection and instance segmentation tasks at various sizes (larger models are slower but can be more accurate). This selection is done in the constructor.

For example:

# initializes a nano-sized instance segmentation model
target_model = YOLOv8("yolov8n-seg.pt")

Available object detection initialization options are:

• yolov8n.pt - nano (3.2M parameters)
• yolov8s.pt - small (11.2M parameters)
• yolov8m.pt - medium (25.9M parameters)
• yolov8l.pt - large (43.7M parameters)
• yolov8x.pt - extra-large (68.2M parameters)

Available instance segmentation initialization options are:

• yolov8n-seg.pt - nano (3.4M parameters)
• yolov8s-seg.pt - small (11.8M parameters)
• yolov8m-seg.pt - medium (27.3M parameters)
• yolov8l-seg.pt - large (46.0M parameters)
• yolov8x-seg.pt - extra-large (71.8M parameters)

License

The code in this repository is licensed under an AGPL 3.0 license.

🏆 Contributing

We love your input! Please see the core Autodistill contributing guide to get started. Thank you 🙏 to all our contributors!

autodistill-yolov8's Issues

device handling in yolov8.py: device=device?

Just a question:
Tried to run the code on windows, and DEVICE overwrote function parameter for self.yolo.train when run with cpu argument on machine with gpu. Is this behaviour intended?

call in main

target_model.train(DATA_YAML_PATH, epochs=2, device = 'cpu')

yolov8.py

DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")

class YOLOv8(DetectionTargetModel):
def init(self, model_name):
self.yolo = YOLO(model_name)

def train(self, dataset_yaml, epochs=300, device=DEVICE):
    print('My sanity', device)
    #self.yolo.train(data=dataset_yaml, epochs=epochs, device=DEVICE)     # CURRENT VERSION 
    self.yolo.train(data=dataset_yaml, epochs=epochs, device=device)         # now takes user definition of defice

Best regards,
Reinhard

Is related to FastSAM ?

Asking out of curiosity. Since the the framework is named with yolov8, by any chance is this work related to FastSAM?