Android application that recognise MNIST Handwritten Digits. This is using a Convolutional Neural Network (CNN) model I made and trained with PyTorch.
This app allows you to recognise images of MNIST Handwritten Digits taken from the official dataset (but not used during model training). You can also draw your own digits and have them recognised by the model.
- You can clone the latest release of this repo using the following command and then open the project in Android Studio. You will then have the possibility to compile the app and launch it on your own Android device or on an emulator.
git clone https://github.com/b-brebion/android-mnist-digits-recognition.git -b v1.5.1
- Or you can directly download the latest APK file available in the Releases tab and run it on your Android device.
CNN(
(conv1): Conv2d(1, 8, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(conv2): Conv2d(8, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(fc1): Linear(in_features=784, out_features=400, bias=True)
(fc2): Linear(in_features=400, out_features=10, bias=True)
(dropout): Dropout(p=0.25, inplace=False)
)
✅ ~99% overall accuracy.
