A Saliency-Based Composition Assistant for Photographers
CompoLens is an AI-powered tool that simulates human visual attention to help photographers assess and improve their image composition. Built for Georgia Tech’s CS 6795: Cognitive Science course, it explores how computational models of saliency can be used to mimic aesthetic decision-making in photography.
Humans tend to focus on specific regions of an image based on visual saliency — features like contrast, brightness, and structure. CompoLens uses state-of-the-art saliency prediction models to:
- Highlight the focal areas of a photograph
- Evaluate how well the intended subject stands out
- Provide real-time feedback to improve framing and balance
- 🖼️ Upload any photo for analysis
- 🧠 Predict saliency maps using AI models (e.g. DeepGaze, SAM, etc.)
- 🎯 Click to indicate your intended focal point
- ✅ Get visual + textual feedback on how well your subject aligns with predicted human attention
- 💡 Built with
Streamlitfor rapid visual prototyping
- Python, Streamlit
- Torch/ONNX (planned support for neural saliency models)
- OpenCV for image processing
- PIL for image manipulation
- Fine-tuning portrait and product composition
- Training aid for photography students and hobbyists
- Creative tool for analyzing balance and framing
- Add more saliency models (DeepGaze, SAM integration)
- Enable drag-and-drop image uploading
- Build batch analysis for contact sheets
- Export annotated images for photographer feedback loops
Coming soon! (Add a screenshot or GIF to show the interface in action)