Object Recognizer

This app performs real time object recognition on data from your phone camera. It utilizes a special visual semantic mapping method for transfer learning from word vectors. This allows it to know many more words for objects than the image model originally learned.

As the full model is installed on the phone, an Internet connection is not required, except to display ads and access the privacy policy.

The current implementation uses a pretrained version of MobileNet in combination with a subset of the FastText word vectors.

Note that coverage of object categories is still somewhat limited right now. We plan to train our own model on a more comprehensive dataset for the next iteration.