### Install Project Dependencies

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Install all necessary Python packages listed in the requirements.txt file. Ensure you have Python 3.8.0 or higher.

```bash
pip install -r requirements.txt
```

--------------------------------

### Python Version Compatibility Error Example

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

This example shows a typical error message encountered when attempting to install dependencies on an incompatible Python version (e.g., 3.7.x).

```text
ERROR: Ignored the following versions that require a different python version: 0.1.0 Requires-Python >=3.9; ...
ERROR: Could not find a version that satisfies the requirement puccinialin
ERROR: No matching distribution found for puccinialin
```

--------------------------------

### Run Face Antispoof Demo (Single Image)

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Use the demo script to analyze a single image for face anti-spoofing detection. Provide the path to the image file.

```bash
python demo.py --image <path>
```

--------------------------------

### Run Face Antispoof Demo (Webcam)

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Execute the demo script to perform real-time face anti-spoofing detection using your webcam. You can specify a camera index if needed.

```bash
python demo.py
# or
python demo.py --camera <index>
```

--------------------------------

### Resume Training

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Continue training from a specific checkpoint. Provide the cropped directory and the path to the checkpoint file.

```bash
python scripts/train.py \
  --crop_dir <path> \
  --resume <checkpoint_path>
```

--------------------------------

### Prepare Face Dataset

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Run the data preparation script to crop, resize, and organize face images for training. Specify original directory, cropped directory, target size, bounding box expansion factor, and spoof types.

```bash
python scripts/prepare_data.py \
  --orig_dir <path> \
  --crop_dir <path> \
  --size <number> \
  --bbox_expansion_factor <float> \
  --spoof_types <number> [<number> ...]
```

--------------------------------

### Create and Activate Virtual Environment (Conda)

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Use Conda to create and activate a new virtual environment for the project. This is recommended for managing dependencies.

```bash
conda create -n face-antispoof python
conda activate face-antispoof
```

--------------------------------

### Create and Activate Virtual Environment (venv)

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Use Python's built-in venv module to create and activate a virtual environment. This is an alternative to Conda.

```bash
python -m venv venv
# Linux/macOS
source venv/bin/activate
# Windows
venv\Scripts\activate
```

--------------------------------

### Prepare Clean Model Weights

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Extract clean, inference-ready model weights from a checkpoint. This removes optimizer states and unnecessary prefixes. Specify the epoch checkpoint, output path, and input size.

```bash
python scripts/prepare_best_model.py <epoch_checkpoint> \
  --output <path> \
  --input_size <number>
```

--------------------------------

### Train Face Anti-Spoofing Model

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Execute the training script using prepared cropped face data. Specify input size, batch size, and output directory for checkpoints and logs.

```bash
python scripts/train.py \
  --crop_dir <path> \
  --input_size <number> \
  --batch_size <number> \
  --output_dir <path>
```

--------------------------------

### Export Model to Quantized ONNX

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Convert a trained model checkpoint to a quantized ONNX format. This can reduce model size and improve inference speed. Specify the checkpoint path, input size, and output path.

```bash
python scripts/quantize_onnx.py <checkpoint_path> \
  --input_size <number> \
  --output <path>
```

--------------------------------

### Export Model to Regular ONNX

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/README.md

Convert a trained model checkpoint to the ONNX format for inference. Provide the checkpoint path, input size, and desired output path.

```bash
python scripts/export_onnx.py <checkpoint_path> \
  --input_size <number> \
  --output <path>
```

--------------------------------

### Image Resizing Interpolation Logic

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/docs/DATA_PREPARATION.md

Selects the interpolation method for resizing based on whether the crop size is smaller or larger than the target size. LANCZOS4 is used for upscaling, and AREA for downscaling.

```python
interp = cv2.INTER_LANCZOS4 if crop_size < target_size else cv2.INTER_AREA
```

--------------------------------

### Temporal Filtering for Video Anti-Spoofing

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/docs/LIMITATIONS.md

This class implements a temporal filter that requires a specified number of consecutive 'real' predictions before accepting a frame as genuine. Use this for video streams to mitigate issues with motion blur and transient spoofing attempts.

```python
from collections import deque

class TemporalFilter:
    """Require N consecutive 'real' predictions before accepting."""
    
    def __init__(self, required_frames=3):
        self.required_frames = required_frames
        self.history = deque(maxlen=required_frames)
    
    def update(self, is_real: bool) -> bool:
        """Update filter and return final decision."""
        self.history.append(is_real)
        
        if len(self.history) < self.required_frames:
            return False  # Not enough frames yet
        
        return all(self.history)  # All must be 'real'
```

--------------------------------

### Crop Face with Padding for Anti-Spoofing

Source: https://github.com/facenox/face-antispoof-onnx/blob/main/docs/LIMITATIONS.md

This function crops a face from an image with a specified padding factor, ensuring sufficient context for the anti-spoofing model. It resizes the cropped face to 128x128 pixels. Use this when the input face bounding box needs to be expanded to include surrounding facial features.

```python
import cv2
import numpy as np

def crop_face_with_padding(image, bbox, padding_factor=1.5):
    """
    Crop face with proper padding for anti-spoofing model.
    
    Args:
        image: Input image (numpy array)
        bbox: Bounding box as (x, y, w, h)
        padding_factor: Padding multiplier (default: 1.5)
    
    Returns:
        Cropped face image
    """
    x, y, w, h = bbox
    
    # Calculate center and expanded dimensions
    center_x = x + w / 2
    center_y = y + h / 2
    max_dim = max(w, h)
    new_size = int(max_dim * padding_factor)
    
    # Calculate new bounding box
    x1 = int(center_x - new_size / 2)
    y1 = int(center_y - new_size / 2)
    x2 = x1 + new_size
    y2 = y1 + new_size
    
    # Clamp to image boundaries
    h_img, w_img = image.shape[:2]
    x1 = max(0, x1)
    y1 = max(0, y1)
    x2 = min(w_img, x2)
    y2 = min(h_img, y2)
    
    # Crop and resize to 128x128
    face_crop = image[y1:y2, x1:x2]
    face_resized = cv2.resize(face_crop, (128, 128), interpolation=cv2.INTER_LANCZOS4)
    
    return face_resized
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.