Smart Crop

A Python tool for batch processing photos that intelligently crops them to any aspect ratio while preserving the most important parts of each image. Perfect for processing large collections of photos, it uses MTCNN for accurate face detection and automatically centers the crop on detected faces, making it ideal for group photos and portraits.

Features

• 📸 Batch process multiple images in parallel
• 👤 Accurate face detection using MTCNN
• 🔄 Automatic EXIF orientation correction
• 📐 Custom aspect ratio support (e.g., 16:9, 4:3)
• ⚡ Multi-processing for faster batch processing
• 💾 Preserves EXIF data in output images
• 🎯 Smart face-based centering
• 🔍 Visual face detection debugging
• ⏱️ Skip existing files to save time

Smart cropping examples

Landscape Example

The orininal photo was cropped to a square, by automatically detecting and centering on the faces and cropping out the left part of the image.

Original	Cropped to square (1:1)

Portrait Example

The orininal photo was cropped to 16:9 ratio (using --ratio 16:9), by automatically detecting and centering on the faces and cropping out most of the top part of the image and also some cropping at the bottom.

Original	Cropped to 16:9

Face Detection Visualization

For the curious, you can turn on visualization of the detected faces by passing --show-faces

Requirements

• Python 3.8 or higher
• Required packages (install with pip install -r requirements.txt):
  • Pillow
  • mtcnn
  • piexif
  • tqdm
  • tensorflow
  • setuptools

Installation

1. Clone this repository
2. Install the required packages:

   pip install -r requirements.txt

Usage

Basic Batch Processing

Simply place your photos in the input directory and run:

python crop.py

Advanced Batch Processing Options

• --input-dir: Input directory containing images (default: "photos")
• --output-dir: Output directory for processed images (default: "output")
• --ratio: Crop ratio in format width:height (e.g., "16:9", "4:3"). Default is square
• --show-faces: Show face detection boxes and confidence scores on output images
• --workers: Number of worker processes (default: number of CPU cores)
• --overwrite: Overwrite existing files in output directory

Batch Processing Examples

Process all images with face detection visualization:

python crop.py --show-faces

Process all images with custom aspect ratio:

python crop.py --ratio 16:9

Process all images with all options:

python crop.py --input-dir my_photos --output-dir processed_photos --ratio 16:9 --show-faces --workers 4 --overwrite

How it Works

Face Detection and Centering

The tool uses MTCNN (Multi-task Cascaded Convolutional Networks) for accurate face detection. When processing an image:

1. Detects all faces in the image with high confidence (95% threshold)
2. For multiple faces:
   • Calculates the center point of all detected faces
   • Centers the crop on this average position
   • Ensures all faces are included in the final crop
3. For single faces:
   • Centers the crop on the detected face
4. If no faces are detected:
   • Falls back to center crop of the image

This intelligent centering ensures that:

• Group photos keep all faces in frame
• Portraits are properly centered on the subject
• Landscape photos maintain their composition when no faces are present

Batch Processing Optimization

• Multi-processing support for parallel batch processing
• Images are automatically resized for faster face detection while maintaining accuracy
• Skips existing files by default to save time on large batches
• Progress bar shows batch processing status
• Efficient memory usage for large photo collections

Directory Structure

.
├── photos/           # Input directory (default)
│   └── *.jpg        # Your photos to process
├── output/          # Output directory (default)
│   └── *.jpg        # Processed photos
├── examples/        # Example images for documentation
├── crop.py
└── requirements.txt

Notes

• The script processes JPG and PNG files in batch
• Face detection confidence threshold is set to 95% to reduce false positives
• Images are automatically resized for faster face detection while maintaining accuracy
• EXIF orientation data is preserved in the output images
• Existing files are skipped by default unless --overwrite is used
• Perfect for processing large collections of photos efficiently