Spaces:
Running
Running
| title: Visionary | |
| emoji: 🚗 | |
| colorFrom: red | |
| colorTo: gray | |
| sdk: static | |
| app_file: index.html | |
| pinned: false | |
| short_description: Car make/model/year recognition, 196 classes, in-browser | |
| # Visionary — fine-grained vehicle recognition (runs in your browser) | |
| Upload a car photo → **YOLOv8n** localises the vehicle → an adaptive **CLAHE** | |
| illumination gate (OpenCV.js) normalises dark or flat images → an INT8 | |
| **ConvNeXt** fine-tuned on Stanford Cars identifies **make, model and year** | |
| (196 classes) with temperature-calibrated confidence. Intermediate classical-CV | |
| stages (CLAHE, bilateral filter, Canny, contours) are shown for every image. | |
| Everything runs locally via ONNX Runtime Web + OpenCV.js; photos never leave the device. | |
| Official test set (8,041 unseen images): **94.1% Top-1 / 99.1% Top-5** (Lite, default) · **94.4 % with Pro** (ConvNeXt-Small @320). | |