from ultralytics import YOLO

# Load model directly from Hugging Face Hub
model = YOLO("s24-farms/apple-detector/best.pt")

# Detect apples in image
results = model("orchard.jpg", conf=0.25, iou=0.45)
for r in results:
    print(f"Detected {{len(r.boxes)}} apples")

Model Details

Property	Value
Architecture	YOLOv8n (nano)
Input size	640×640
Confidence threshold	0.25
NMS IoU	0.45
Training images	200 synthetic + 3000 augmented
Classes	1 (`apple`)

Detection Pipeline

YOLOv8n — Primary detector (fast, ~0.2s per image on CPU)
OpenCV Color Blob — Fallback for red/yellow apples
Watershed Segmentation — Separates touching apples
Hough Circles — Detects round shapes with apple-like color

Features per Apple

Bounding box with confidence score
Estimated diameter (mm) with range (±5mm)
Estimated weight (g) with range (±15%)
Variety classification (Red/Golden/Granny/Mixed)
Disease detection (Scab, Bitter Rot, Sooty Blotch, Sunburn, Powdery Mildew)
Size classification (Small/Medium/Large)

Video Tracking (ByteTrack)

python

from tracker import process_video_file

result = process_video_file("orchard_row.mp4", max_frames=300)
print(f"Unique apples: {{result['summary']['total_unique_apples']}}")

Web API (FastAPI)

bash

pip install -r requirements.txt
python api_fastapi.py 8080
# → http://localhost:8080/docs

Dataset

The model was trained on 200 synthetic apple images (generated in Unreal Engine) and augmented to 3200 images using Albumentations:

Random brightness/contrast (±30%)
HSV shifts (H±20, S±30, V±20)
Gaussian/Motion blur
CLAHE, GaussNoise, RandomShadow
Horizontal flip

Citation

@software{s24_farms_apple_detector,
  author = {{S24 Farms}},
  title = {{S24 Farms Apple Detection}},
  year = {2026},
  url = {https://huggingface.co/s24-farms/apple-detector}
}

Shadyemad/s24-apple-detector

🍎 S24 Farms Apple Detection Model

Usage

Model Details

Detection Pipeline

Features per Apple

Video Tracking (ByteTrack)

Web API (FastAPI)

Dataset

Citation

Shadyemad/
s24-apple-detector