---

🎯 Demo Objectives

The goal of this demo is to demonstrate a complete and practical computer vision pipeline, including:

• Multi-object fashion detection
• Improved recall for small objects (rings, watches, accessories)
• Multi-label attribute prediction per detected object
• Clean, standardized output format for downstream usage

This demo is intended as a proof-of-concept system.

---

🚀 Features

✅ Object Detection

• YOLOv11-based detection
• Supports apparel and accessory categories
• GPU acceleration via CUDA when available

✅ SAHI Integration

• Optional SAHI inference for small objects
• Toggleable directly from the UI
• Helps recover tiny fashion items often missed by standard inference

✅ Attribute Recognition

• Each detected object is cropped and passed to an attribute head
• Multi-label prediction (one object → multiple attributes)
• Attribute IDs are mapped to human-readable names

✅ Output Formats

• Visual output with bounding boxes
• Tabular summary (class, confidence, attributes)
• JSON output matching submission format

Example JSON output:

[
  {
    "label": "Cardigan",
    "confidence": 0.97,
    "box": [100, 175, 715, 971],
    "attributes": ["Plain pattern", "Short length", "Single breasted"]
  }
]