This repository houses the source code for the UVA Landmark Recognition App, currently available on iOS via the App Store and on Android via the Google Play Store.
This app has two main use cases:
- acquiring data for use in training a computer vision model
- allowing users to test such a model using their camera or image library
For data storage, this app uses Firebase. Firebase project keys can be specified in FirebaseConfig.js, and all relevant code used in the collection screen can be found and modified in Screens/CollectionScreen.js. Landmark category names can be specified in SiteNames.js.
Once the prediction screen is initialized (Screens/PredictionScreen.js), a model is downloaded from a pre-defined URL and loaded using tf.js. Afterwards, users are able to provide an image by either selecting one from their device's image library or taking a photo using their device's camera. This image is then converted into a tensor and appropriately pre-processed to allow for prediction to occur. Upon prediction, the app displays the top-3 prediction results along with their confidence metrics.
A walkthrough of the machine learning side to this project can be found at https://github.com/sg2nq/UVA-Landmark-Recognition.