EdgeFire-Lab Models

Automatically trained and versioned YOLOv8 models from the EdgeFire Lab Pipeline.

Overview

This repository contains YOLOv8 models trained for [deine Anwendung - z.B. wildfire detection, object detection, etc.].

Models are automatically uploaded after training and versioned by run ID.

Available Models

Models will appear here after first upload.

Quick Start

Installation

pip install huggingface_hub ultralytics
Download and Use a Model

from huggingface_hub import hf_hub_download
from ultralytics import YOLO

# Download specific model
model_path = hf_hub_download(
    repo_id="weltraumlama/yolov8-edgefire",
    filename="models/run_001/weights/best.pt"
)

# Load model
model = YOLO(model_path)

# Run inference
results = model.predict("image.jpg")

# Process results
for result in results:
    boxes = result.boxes  # Bounding boxes
    print(boxes.xyxy)     # Box coordinates
    print(boxes.conf)     # Confidence scores
    print(boxes.cls)      # Class labels

Model Structure Each model directory contains:

• weights/best.pt - Best checkpoint (lowest validation loss)
• weights/last.pt - Last epoch checkpoint
• results.csv - Training metrics per epoch
• args.yaml - Training hyperparameters
• *.png - Visualization plots (confusion matrix, PR curve, etc.)

Training Pipeline Models are trained using the EdgeFire Lab Pipeline:

• Preprocessing - Image preprocessing and augmentation
• Labeling - Annotation and quality control
• Consolidation - Dataset merging and validation
• Dataset Split - Train/val/test splitting with markers
• Training & Inference - YOLOv8 training with auto-promotion

Training Configuration

• Framework: Ultralytics YOLOv8
• Auto-promotion: Best models are automatically marked as production
• Marker-based: Prevents duplicate training runs