Detect and locate objects in images. Bounding boxes with confidence scores. Runs entirely in your browser using DETR ResNet-50 — private, no upload.
Your images are processed entirely in your browser via DETR ResNet-50. Nothing is uploaded to any server.
Drop image here or click to select
PNG, JPG, WebP — single image
Runs entirely in your browser. Your data never leaves your device.
Tested on 100 COCO validation images — average detection time 3.2s on MacBook Air M2.
| Metric | Value |
|---|---|
| Model size (INT8) | ~100MB |
| COCO mAP | 42.0 (ResNet-50 backbone) |
| Avg detection time | 3.2s / image |
| Model load (cold start) | ~18s (CDN) / <1s (cached) |
| Detectable categories | 80 (COCO) |
Last verified: September 2026.
No. This AI object detector runs entirely in your browser using WebAssembly. Images are processed locally via DETR ResNet-50 and never leave your device.
DETR ResNet-50 is trained on COCO and can detect 80 object categories including person, car, chair, dog, cat, book, bottle, sports ball, and many more common objects.
A bounding box is a rectangular box that indicates the location of a detected object. Each box comes with a class label and a confidence score (0-100%) showing how sure the model is.
Use images with clear, well-lit objects. Adjust the confidence threshold slider to filter out low-confidence detections. Lower threshold = more detections; higher = fewer but more accurate.