# 📋 Overview

This repository contains optimized AI models for **XAIVA Kiosk Analysis System** - an intelligent workplace safety and compliance monitoring solution. The models are specifically designed for real-time analysis in industrial environments.

## 🚀 Model Collection

### 🎯 Detection Models
- **`best_0327.engine`** (131MB)
  - **Type**: Object Detection (TensorRT Optimized)
  - **Purpose**: Safety equipment detection (caps, masks, gloves, boots, etc.)
  - **Input**: RGB images
  - **Output**: Bounding boxes with class predictions

### 👤 Human Analysis Models
- **`yolo11n-seg.engine`** (10.2MB)
  - **Type**: Instance Segmentation (TensorRT Optimized)  
  - **Purpose**: Person segmentation for workspace analysis
  - **Architecture**: YOLOv11 Nano
  - **Output**: Pixel-level person masks

- **`yolo11n-pose.engine`** (10.1MB)
  - **Type**: Pose Estimation (TensorRT Optimized)
  - **Purpose**: Human pose detection for safety compliance
  - **Architecture**: YOLOv11 Nano
  - **Output**: 17-point keypoint coordinates

### 🔍 Face Recognition Models
- **`facedet_fp16.trt`** (10.3MB)
  - **Type**: Face Detection (TensorRT FP16)
  - **Purpose**: Real-time face detection in workplace
  - **Optimization**: Half-precision for speed
  - **Output**: Face bounding boxes with landmarks

- **`adaface_ir101.trt`** (132MB)
  - **Type**: Face Recognition (TensorRT Optimized)
  - **Purpose**: Employee identification and access control
  - **Architecture**: AdaFace with ResNet-101 backbone
  - **Output**: 512-dimensional face embeddings


## 🛠️ Usage

### Prerequisites
```bash
pip install torch torchvision
pip install ultralytics
pip install opencv-python
pip install tensorrt
```

## 🏗️ System Requirements

- **GPU**: NVIDIA GPU with Compute Capability 6.0+
- **VRAM**: 8GB+ recommended
- **CUDA**: 11.0+
- **TensorRT**: 8.0+
- **Python**: 3.8+

## 📁 Model Files

```
models/
├── best_0327.engine          # Main detection model
├── yolo11n-seg.engine        # Person segmentation
├── yolo11n-pose.engine       # Pose estimation
├── facedet_fp16.trt          # Face detection
└── adaface_ir101.trt         # Face recognition
```

## 🔧 Integration

These models are designed to work together as part of the XAIVA Kiosk system:

1. **Face Detection** → **Face Recognition** for employee identification
2. **Detection** → Safety equipment compliance checking
3. **Segmentation** → Workspace occupancy analysis  
4. **Pose Estimation** → Safety posture monitoring

## 🏢 About XIILAB

Developed by [XIILAB](https://xiilab.com/main) - Advanced AI solutions for industrial automation and safety.