⏱ Duration: Jan 25 – Apr 25
🛠 Tech Stack: Python, PyTorch, React.js, Django, Tailwind CSS
🔗 GitHub: GitHub Repo
🔗 Website link: Website Link
Manual fashion tagging is a billion-dollar bottleneck for fashion e-commerce platforms. With ever-expanding product catalogs, tagging new items accurately and consistently has become unscalable.
-
With manual input taking around one minute per product, scaling to millions of products in a seamless flow poses a significant operational challenge.
Manual tagging struggles with:
-
🐌 Speed Bottleneck: Products can take days to become searchable due to 10× slower throughput.
-
📈 Scalability Nightmare: Tagging needs grow linearly with SKUs, causing unsustainable costs.
A robust, full-stack system to auto-tag fashion product images using a deep learning pipeline, with real-time predictions and downloadable reports.
- ✅ Multi-label classification (5 - 10 fashion attributes on 5 categories of clothing)
- 🔁 Transfer learning with domain-specific tuning
| Metric | Value |
|---|---|
| ⚡ Inference Time | < 1s |
| 🏷 Attributes Handled | 5 - 10 attributes for 5 clothing categories |
| 🎯 Evaluation Metric | 73% harmonic F1 |
Frontend (React.js + Tailwind) -> Backend (Django REST API) -> Model Server (PyTorch) -> Attribute Output (JSON, CSV, Reports)