Image Sorter

An automated image categorization tool that uses deep learning to sort images into predefined categories.

Overview

This application uses a ResNet50-based neural network to classify and sort images into categories. It works in two phases:

1. Training Phase: The model learns from categorized images in the dataset directory
2. Sorting Phase: The trained model automatically sorts images from myImages into appropriate categories in the finalSorted directory

Features

• Uses transfer learning with pre-trained ResNet50 model
• Automatically creates category folders in the destination directory
• Preserves original images (copies rather than moves)
• Supports .jpg, .jpeg, and .png file formats
• Configurable via environment variables or command line arguments
• Saves the trained model for future use without retraining

Requirements

• Python 3.6+
• Dependencies listed in requirements.txt:

  tensorflow>=2.4.0
  scikit-learn>=0.24.0
  pillow>=8.0.0
  numpy>=1.19.0
  python-dotenv>=0.19.0
  

Directory Structure

├── dataset/           # Training data (organized by category)
│   ├── category1/     # Each subfolder represents a category
│   │   ├── img1.jpg
│   │   ├── img2.jpg
│   ├── category2/
│   │   ├── img3.jpg
│   │   └── ...
├── myImages/          # Images to be sorted
│   ├── image1.jpg
│   ├── image2.png
│   └── ...
├── finalSorted/       # Destination for sorted images (created automatically)
│   ├── category1/
│   ├── category2/
│   └── ...
├── model/             # Saved model and metadata (created automatically)
│   ├── image_sorter.h5
│   ├── label_encoder.pkl
│   └── categories.pkl
├── .env               # Environment variables (optional)
├── main.py            # Original script
└── image_sorting_reuse.py  # Enhanced version with model saving/loading

Usage

Basic Usage

1. Prepare your dataset:

   • Create a dataset directory
   • Inside it, create subdirectories for each category (e.g., dogs, cats, landscapes)
   • Place sample images in each category directory

2. Prepare images to sort:

   • Place images you want to sort in the myImages directory

3. Run the script:

   python image_sorting_reuse.py
   

4. Find sorted images:

   • Check the finalSorted directory for your categorized images

Command Line Arguments

The enhanced version (image_sorting_reuse.py) supports these arguments:

• --train: Force training a new model even if one already exists
• --input [path]: Specify a custom directory containing images to sort (default: ./myImages)

Example:

python image_sorting_reuse.py --train --input ./vacation_photos

Configuration

You can customize the application by setting these environment variables in a .env file:

DATASET_DIR=./dataset
FINAL_SORTED_DIR=./finalSorted
MY_IMAGES_DIR=./myImages
MODEL_PATH=./model/image_sorter.h5
LABEL_ENCODER_PATH=./model/label_encoder.pkl
CATEGORIES_PATH=./model/categories.pkl
BATCH_SIZE=32
EPOCHS=10
VALIDATION_SPLIT=0.2
LEARNING_RATE=0.001

How It Works

1. Feature Extraction: Uses ResNet50 (pre-trained on ImageNet) to extract high-level features from images
2. Model Training: Trains a classifier on top of these features to recognize your specific categories
3. Prediction: Applies the trained model to new images and sorts them based on predictions
4. Persistence: Saves the trained model and metadata for future use

Performance Considerations

• Dataset Quality: Include diverse, high-quality examples for each category
• Dataset Size: Aim for at least 20-30 images per category for decent results
• Training Time: Depends on dataset size and hardware
• Accuracy: Heavily dependent on how distinct your categories are and dataset quality

Troubleshooting

• Incorrect Classifications: Add more training examples to the category that's being misclassified
• Model Loading Errors: Delete the model directory and retrain by using the --train flag
• Memory Errors: Reduce BATCH_SIZE in the .env file

Contributing

Feel free to submit issues or pull requests to improve this project.

License

This project is open source and available under the MIT License.