👗 FaSHioN

Editorial-Grade Fashion Image Clustering & Visual Discovery

FaSHioN is an unsupervised fashion image clustering pipeline designed to organize, explore, and discover visual patterns in fashion imagery at an editorial and catalog scale. By leveraging deep visual embeddings and classical machine learning, FaSHioN enables automated grouping of fashion images based purely on visual similarity—without labels.

Live Demo: https://fashion-clustering.streamlit.app/

• Model performance in the Streamlit application may be limited, as the model does not have access to the full training dataset during deployment. Instead, predictions rely on individually uploaded images via GitHub, which can affect overall accuracy and consistency.

---

✨ Key Features

• Deep Visual Understanding using pre-trained ResNet50
• Unsupervised Clustering with K-Means
• Dimensionality Reduction via PCA
• Editorial-grade visual discovery for trend analysis, styling, and catalog curation

---

🧠 How It Works

1. Feature Extraction
   Images are passed through a pre-trained ResNet50 (ImageNet) to extract high-level visual embeddings.
2. Dimensionality Reduction
   High-dimensional embeddings are compressed using Principal Component Analysis (PCA) to improve clustering efficiency and separability.
3. Clustering
   K-Means groups images into visually coherent clusters representing styles and silhouettes.
4. Visual Exploration
   Clustered outputs can be used for visual inspection, similarity search, or downstream fashion analytics.

---

📦 Tech Stack

• Python
• TensorFlow / Keras – ResNet50
• Scikit-learn – PCA & K-Means
• NumPy / Pandas
• Matplotlib / Seaborn (optional visualization)

🔮 Future Enhancements

• Color-aware clustering
• Use self-supervised vision models (CLIP, DINO)