Retour aux projets
Real-time object detection
A browser-native demo powered by COCO-SSD (90 classes) and TensorFlow.js. Everything runs locally: upload a photo or enable your webcam and watch detections appear instantly.
Erreur de chargement du modèle
Ou essayez avec une image d'exemple:
Tech stack
- •TensorFlow.js – runs the COCO-SSD model fully client-side
- •React Hooks – manage camera/video lifecycle and inference loop
- •Canvas API – draws bounding boxes and labels on top of the video feed
- •TypeScript – typed detection pipeline for more robust UI state
Highlights
- ✓Privacy first: 100% client-side, no uploads to any server.
- ✓Real time: optimized inference loop for smooth webcam usage.
- ✓90+ classes: people, animals, vehicles, everyday objects.
- ✓Confidence slider: tune sensitivity for your use case.
- ✓Responsive: works on desktop and mobile browsers.
⚡️
Performance note
The model loads on first use (a few seconds). Once cached, detections are near-instant. For the best FPS, keep uploads under ~2 MB or reduce webcam resolution.
Use cases
🛒
E-commerce
Auto-tagging products, visual inventory control, image-based search.
🏭
Industrial
Automated quality checks, anomaly detection, on-the-fly item counting.
🔐
Security
Smart surveillance, intrusion detection, and lightweight video analytics.