Android app that detects handwritten digits.
The app uses a sequential TensorFlow model and the structure of the network is as follows.
- Convolution2D layer with activation 'relu' with 3 filters of size 3 and input_shape 28x28x1 output shape 32
- Convolution2D layer with activation 'relu' and 3 filters of size 3
- MaxPooling2D layer of pool_size (2,2)
- Dropout layer of size 0.25
- Flatten layer
- Dense layer of size 128 with activation 'relu'
- Dropout layer of size 0.5
- Output Dense layer of size 10 with activation 'softmax'
The app uses TensorFlow Lite Interpreter in Android Studio. The model is Interpreted from the tflite file for on device inferencing.