Sharpness Indexer

Sharpness Indexer was born out of a need to save time after motorsport photography sessions. When capturing panning shots—where you follow a moving car and take bursts of images—you often end up with hundreds of pictures, many of which are blurry or imperfect due to motion. Reviewing each photo manually to find the sharp ones, where the car is properly captured, can be extremely time-consuming. This script was created to automate that process.

This tool is written in Python and is designed to analyze the sharpness of cars in high-resolution images, even when the background is blurry—as is typical in panning shots. It uses YOLOv8 to detect cars and Laplacian variance to evaluate image sharpness. Once processed, each image is renamed with a prepended sharpness value, and all images are saved to a single output folder where you can then simply sort the files by filename descending and have the sharpest pictures at the top.

---

🚀 Features

• ✅ Uses YOLOv8 to detect objects (e.g., cars, people, etc.)
• ✅ Measures sharpness (Laplacian variance) of each detected object
• ✅ Prepends the highest sharpness score to each filename (e.g., S134__IMG001.jpg)
• ✅ Copies or moves processed images to a single output folder
• ✅ Supports multiple image formats (.jpg, .jpeg, .png)
• ✅ Utilizes GPU acceleration if available (via PyTorch + CUDA)

---

🖼️ Example

Original filename	Detected sharpness	Output filename
IMG_001.jpg	134.23	S134__IMG_001.jpg
trackshot.png	41.87	S041__trackshot.png

---

📦 Installation

1. Clone the repository

   git clone https://github.com/yourusername/sharpness-indexer.git
   cd sharpness-indexer

2. Create a virtual environment

   python3 -m venv yolo

3. Activate the virtual environment

   • On Windows:

     yolo\Scripts\activate

   • On macOS/Linux:

     source yolo/bin/activate

4. Install dependencies

   Python 3.9+ and pip required

   pip install -r yolo-requirements.txt

5. Optional: Check for GPU availability

   import torch
   print(torch.cuda.is_available())
   print(torch.cuda.get_device_name(0))

---

📁 Folder Structure

sharpness-indexer/
├── input/                   # Put your input images here
├── output/                  # Processed images will appear here
├── README.md                # This file
├── sort-yolo.py             # Main script
├── yolo-requirements.txt    # Requirements for `sort-yolo.py`

---

🧠 How It Works

1. You choose a YOLOv8 model (n, s, m, l, x) – defaults to x (yolov8x.pt).
2. The script loads each image from the input/ folder.
3. YOLOv8 detects all objects in the image (not limited to cars).
4. Sharpness is calculated for each object using the Laplacian operator.
5. The highest sharpness score among all detected objects is used.
6. The image is renamed using the format: S{SHARPNESS}__{ORIGINAL_NAME}.
7. Depending on your selection, the image is either copied or moved to the output/ folder.

---

⚙️ Configuration

Setting	Description	Default
YOLOv8 model	Detection backbone (n, s, m, l, x)	yolov8x.pt
Sharpness metric	Laplacian variance (computed via OpenCV)	-
Output format	S{SHARPNESS}__{ORIGINALNAME}.{EXT}	e.g. S098__IMG1.jpg
Input directory	Folder containing raw images	input/
Output directory	Folder for renamed files	output/

---

📌 Notes

• All detected objects (not just cars) are evaluated for sharpness.
• Sharpness is computed using Laplacian variance via OpenCV.
• If no object is found, the sharpness is set to 0.00.
• Original images remain untouched unless you choose to move them.

---

🛠️ Requirements

• Python 3.9+
• PyTorch (with CUDA for GPU support)
• Ultralytics YOLOv8
• OpenCV

---

✅ Example Run

python3 sort-yolo.py

Choose YOLOv8 model (n = nano, s = small, m = medium, l = large, x = xlarge)
Model (default: x):
[INFO] Using model: yolov8x.pt
[OK] IMG_001.jpg -> S134__IMG_001.jpg
[OK] IMG_002.jpg -> S023__IMG_002.jpg

---

📖 License

MIT License © 2025

---

💬 Questions?

Feel free to open an issue if you have suggestions or questions!

---