With this app you can playfully test the ability of a neural network to classify objects using pictures. You can take pictures of different classes of objects and train a neural network with the collected data set. After uploading the data set, the training takes place on powerful servers in the cloud. After a few minutes, the trained network is downloaded on to your device and you can try out live how well the image recognition works. This takes place exclusively on your device and not in the cloud.
For example, you can create a class cup and a class fork and add images of these objects. At least 10 images per class are needed to train the neural network. To get the best recognition rate, take pictures of as many objects of a category as possible. It also helps to take pictures of objects from different perspectives.
Classes can be added, deleted or extended with new images afterwards. Dont forget to retrain the neural network after the changes. You can also share datasets with other users of the app via a link.
The neural network is based on the MobileNet V2 architecture and is trained using TensorFlow.