Flicked Fashion AI is a smart computer vision pipeline that analyzes Instagram-style short videos, detects fashion items worn by people, matches them to a product catalog, and classifies the outfit's overall aesthetic vibe.
- 🎯 YOLOv8n — detects fashion products (tops, trousers, dresses) frame by frame
- 🧠 CLIP (ViT-B/32) — embeds both catalog images and detected crops into 512-d vectors
- ⚡ FAISS — finds exact/similar matches from the catalog at scale
- 🎨 NLP Vibe Classifier — assigns fashion aesthetics like Y2K, Clean Girl, Streetwear from product metadata
- 📦 Structured JSON output — production-ready for ecommerce, tagging, or marketing pipelines
Video → Frame Extraction → YOLOv8 Detection → CLIP Embedding → FAISS Matching → JSON Output → Vibe Classification
- Frame Extraction — OpenCV extracts 1 frame/sec from each reel in
/videos/reel/ - Fashion Detection — YOLOv8n detects and crops items → saved to
data/cropped_items/ - Catalog Embedding — CLIP encodes catalog images from
catalog.csv→ FAISS index built once and reused - Matching — Each crop is embedded and compared to catalog; top-1 match selected if similarity > 0.75
- Per-Video Output — Items grouped by reel → saved as
/outputs/reel_XXX.json
- Vibe Classification — Keywords extracted from title/description/tags → matched to
vibes_list.json
git clone cd submission/ pip install -r requirements.txt python embed_catalog.py python detect_and_crop.py python match_with_catalog.py python classify_vibes.py
- Catalog embeddings are built once and reused — no redundant computation
- Vibe classifier uses natural language from real product data, not just class labels
- JSON outputs are production-ready for downstream ecommerce or marketing use
- Designed for scalability with plans for a product-level REST API
- Browser extension / Instagram plug-in — tap any outfit in a reel → get buy links
- Upload 2–3 photos → get smart outfit suggestions with vibes + live ecommerce results
- Brand analytics — analyze UGC trends in real time
Built by Vinayak