About Med Herb Lens

Identify medicinal herbs with your phone's camera, browse 4,799 plant species with their Vietnamese and English names, and train image classifiers of your own, all on your device. Med Herb Lens runs on Android, iOS and the web.

A research preview. Med Herb Lens is a research project. Its herb model is our own: it was trained by our team, runs on your device and knows 2,721 species, but we don't have enough photos yet for most of them, so its results are often wrong. No other company's identification service is used. Never eat a plant or use it as medicine because of what the app says.

Get the app

In your browser: med-herb-lens.pages.dev/app
Android: Google Play
iPhone and iPad: App Store

What it does

Identify. Point the camera at a plant or pick photos; the app suggests which species it may be, with its confidence. Whole view identifies everything in sight; Pick a plant finds each plant and identifies one at a time. Identification happens on the device and works offline.
Browse and search. 4,799 species from the GBIF backbone taxonomy, searchable by Vietnamese (with or without diacritics), English or scientific name. Each species shows its names, classification and full GBIF record, with reference photos from GBIF and iNaturalist contributors and photos shared by users.
Saved. Favourites and the herbs you viewed recently.
Contribute. Signed-in users share photos of herbs (with where they were taken) and suggest common names (Vietnamese or English for now). Shared photos can be reported, and a contributor's photos hidden.
Train your own model. Collect photos of the plants you care about, train a classifier on the device and try it with the camera. Share it as a standard .tflite file that carries its own species list, so it works in any app that runs TensorFlow Lite models; imported into Med Herb Lens on another device, it can go on learning there. You can also share a model with everyone in the app.
Research mode. Run continual-learning experiments on the device: choose a dataset, scenarios, strategies and seeds, leave the device running (even with the screen off), then save one archive with every result, the device's measurements and the trained models.
How to cite. The papers behind the app, in Profile, ready to copy.

Privacy

Photos are identified on your device and leave it only when you choose to share them. The app sends anonymous usage statistics (such as which herbs are identified) for research, with no name, email, photo or place; you can turn them off in Profile. No ads, no tracking. Details in the privacy policy.

Research and how to cite

If you use Med Herb Lens, its data or its code in your work, please cite:

Tran, T. P., Ud Din, F., Brankovic, L., Sanin, C., & Hester, S. M. (2025). Med Herb Lens: A prototype AI app for medicinal plant identification. Procedia Computer Science, 270, 2603–2612. https://doi.org/10.1016/j.procs.2025.09.382

Tran, T. P., Ud Din, F., Brankovic, L., Sanin, C., & Hester, S. M. (2026). Resource-efficient continual learning for medicinal plant identification: A periodic retraining approach for edge-deployed agricultural IoT applications. IoT, 7(3), 57. https://doi.org/10.3390/iot7030057

Built with

Kotlin Multiplatform and Compose Multiplatform: one app for Android, iOS and the web
LiteRT (TensorFlow Lite) for on-device identification, with MediaPipe image embedders as the backbones of user-trained models; their final layers train in shared Kotlin
Firebase: sign-in, shared photos and name suggestions, usage statistics, remote settings and App Check
Cloudflare: the website and web app (Pages), photo and model storage (R2) and uploads (Workers)
GBIF: the species catalog and reference photos

License

The code is licensed under the Apache License 2.0. The herb identification model, the species catalog as compiled for the app, the documentation and website text, and images are licensed under the Creative Commons Attribution 4.0 International licence (CC BY 4.0): you may share and adapt them for any purpose, as long as you give credit (see Research and how to cite). NOTICE says which is which.

Others' work the app includes keeps its own terms: the catalog draws on GBIF data, reference photos remain their creators' (each shown with its licence, many for non-commercial use only), the MediaPipe embedders are under Apache 2.0, and each library under its own licence. Models people share in the app are theirs, shared under CC BY 4.0.

Contact

Trien Phat Tran, University of New England: ttran72@myune.edu.au or tptrien@gmail.com