A comprehensive computer vision system for analyzing and evaluating photos using state-of-the-art deep learning models.
- Facial Analysis: Detection, recognition, expression analysis, and face grouping
- Aesthetic Scoring: Visual quality assessment and print-worthiness evaluation
- Semantic Tagging: CLIP-based tagging and BLIP image captioning
- Similarity Search: DINOv2 embeddings with FAISS vector search
- Scene Understanding: Object detection and scene classification
# Install dependencies
pip install -r requirements.txt
# Analyze a photo
python scripts/demo_photo_eval.py path/to/photo.jpg
# Batch process
python scripts/demo_photo_eval.py path/to/photos/ --batch --group-facesFull pipeline run on four freely-licensed sample images — reproduce with
python scripts/generate_sample_outputs.py:
| Image | Caption (BLIP) | Scene (CLIP) | Aesthetic | Faces | Top tags (CLIP) |
|---|---|---|---|---|---|
| portrait | "a woman in an orange space suit with a space shuttle in the background" | open sky (0.76) | 62.9/100 | 1 | bright, park, woman, sunny, bird |
| pet | "a close up of a cat's face with a blurry background" | park or garden (0.15) | 51.3/100 | 0 | cat, animal, sunny, bright, cloudy |
| food | "a cup of coffee on a saucer with a spoon on a table" | restaurant or cafe (0.55) | 52.6/100 | 0 | dark, food, sitting, standing, day |
| scene | "a spacex falcon rocket taking off from the launch pad" | open sky (0.85) | 52.4/100 | 0 | night, day, bright, person, people |
The portrait yields a real face detection (confidence 1.0) plus a 512-d Facenet512 embedding and age/gender/expression attributes; non-portrait images correctly report 0 faces. See PHOTO_EVALUATION_README.md for details.
- PHOTO_EVALUATION_README.md - Complete documentation
- QUICKSTART.md - Quick reference guide
PhotoEvaluationAgent
├── FacialAnalysisAgent (DeepFace)
├── AestheticAnalysisAgent (CLIP + LAION aesthetic head)
├── SemanticAnalysisAgent (CLIP + BLIP)
├── SimilarityAnalysisAgent (DINOv2 + FAISS)
└── SceneDetectionAgent (DETR + CLIP zero-shot)
- Python 3.8+
- PyTorch 2.0+
- See requirements.txt for full list
MIT License - see LICENSE file for details