Lab · Licensing
In-browser object detection
Recommendation
Use TensorFlow.js COCO-SSD (Apache-2.0) for the public personal-site demo.
- Runs fully in the visitor’s browser (WebGL/CPU); no server GPU.
- Apache-2.0 is permissive: fine for a personal site without AGPL copyleft on the whole project.
- Mature CDN distribution (
@tensorflow/tfjs+@tensorflow-models/coco-ssd). - Good enough for a teaching sample (COCO classes); not a claim of “Ultralytics YOLO production accuracy.”
Name the page honestly: “object detection sample (COCO-SSD)” while the podcast can still explain the YOLO idea. Do not imply the lab page loads Ultralytics weights if it does not.
Comparison
Verify again before any production change.
| Option | License | Browser-friendly? | Fit for public personal site |
|---|---|---|---|
| Ultralytics YOLOv5/v8/11 | AGPL-3.0 (Enterprise for closed use) | Possible via exports, but AGPL follows | Risky unless whole derivative is AGPL’d or Enterprise licensed |
| YOLOX | Apache-2.0 | Needs your own ONNX/TFJS path | Good license; more build work |
| TF.js COCO-SSD | Apache-2.0 | Yes (official) | Recommended for this lab |
| MediaPipe Object Detector | Apache-2.0 (SDK); confirm each model bundle | Yes (WASM) | Strong alternative |
| RT-DETR variants | Varies (Ultralytics port → AGPL) | Uneven | Only with a clearly permissive browser build |
Sources
- Ultralytics License (AGPL vs Enterprise)
- Ultralytics Terms §5
- Ultralytics pricing (Free/Pro = AGPL)
- TF.js coco-ssd README
- npm @tensorflow-models/coco-ssd (Apache-2.0)
- YOLOX (Apache-2.0)
- MediaPipe LICENSE
- MediaPipe Tasks Vision
CDN libraries used by this lab
@tensorflow/tfjs@4.22.0via jsDelivr@tensorflow-models/coco-ssd@2.2.3via jsDelivr- Default weights: SSD Lite MobileNet v2 via TensorFlow.js model hosting
Retain Apache notices; do not strip license headers if you later vendor instead of CDN.