README.md

metadata

library_name: ultralytics
tags:
  - yolov11
  - object-detection
  - instance-segmentation
  - computer-vision
  - deep-learning
  - port-detection
license: agpl-3.0

Port Model

This is a custom trained YOLOv11 segmentation model for port detection.

Model Details

• Model Type: YOLOv11 Instance Segmentation
• Framework: Ultralytics YOLOv11
• Task: Instance Segmentation
• Classes: 2
• Input Size: 1408x1408
• Dataset: Custom Port Dataset

Classes

• Class 0: Port-capped
• Class 1: Port-Empty

Model Configuration

{
  "model_type": "yolov11-seg",
  "task": "image-segmentation",
  "framework": "ultralytics",
  "num_classes": 2,
  "id2label": {
    "0": "Port-capped",
    "1": "Port-Empty"
  },
  "input_size": 1408,
  "confidence_threshold": 0.25,
  "iou_threshold": 0.45
}

Training Configuration

• Epochs: 100
• Batch Size: 16
• Optimizer: AdamW
• Dataset: Custom Port Dataset

Usage

Using Ultralytics (Local Inference)

from ultralytics import YOLO

# Load model
model = YOLO('model.pt')

# Run inference
results = model('image.jpg', conf=0.25, iou=0.45)

# Process results
for result in results:
    masks = result.masks  # Segmentation masks
    boxes = result.boxes  # Bounding boxes
    
    # Get class names
    for box in boxes:
        class_id = int(box.cls)
        class_name = {"0": "Port-capped", "1": "Port-Empty"}[str(class_id)]
        confidence = float(box.conf)
        print(f"Detected: {class_name} ({confidence:.2f})")
    
    # Visualize
    result.show()

Using Hugging Face Inference API

import requests
import json

API_URL = "https://router.huggingface.co/models/Sunix2026/Port-model"
headers = {"Authorization": "Bearer YOUR_HF_TOKEN"}

def query(filename):
    with open(filename, "rb") as f:
        data = f.read()
    response = requests.post(API_URL, headers=headers, data=data)
    return response.json()

# Run inference
output = query("image.jpg")
print(json.dumps(output, indent=2))

Using the Python Client

from yolov11_hf_inference import YOLOv11HFInference

# Initialize client
client = YOLOv11HFInference(
    model_url="Sunix2026/Port-model",
    access_token="YOUR_HF_TOKEN"
)

# Run inference
result = client.predict_from_path("image.jpg")

if result["success"]:
    predictions = result["predictions"]
    
    # Map class IDs to names
    id2label = {"0": "Port-capped", "1": "Port-Empty"}
    
    for pred in predictions:
        class_name = id2label.get(str(pred.get('label', '')), 'Unknown')
        confidence = pred.get('score', 0)
        print(f"Found: {class_name} ({confidence:.2%})")
else:
    print(f"Error: {result['error']}")

Performance Metrics

Metric	Value
Confidence Threshold	0.25
IoU Threshold	0.45
Input Resolution	1408x1408

Applications

This model can be used for:

• Port detection and classification
• Automated quality control
• Manufacturing inspection
• Inventory management

Limitations

• Model is trained specifically for port detection
• Performance may vary with different lighting conditions
• Best results with images similar to training data

License

AGPL-3.0

Citation

If you use this model, please cite:

@misc{Port-model,
  author = {Sunix2026},
  title = {Port Model},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Sunix2026/Port-model}}
}