Turn your mouse movements into text.
The app uses a LSTM that classifies mouse movements read from the combined device file for all mice (/dev/input/mice).
It can work with both udev and eudev device filesystems.
The model has only been trained on with a mouse set at a dpi of 800. The dataset contains about 1000 training samples, and it is not balanced. Letter which were difficult for the model to classify have more samples than easier letters.
The user can draw the letters using the mouse movements associated with the respective letters.
The user needs to be in the input group to be able to read device files.
sudo usermod -aG input $USER
A restart is needed for these changes to take effect.
-
dataset.json: A dataset for testing whether model can recognise different letters. Too small with similar letters. -
dataset2.json: A dataset with carefully chosen shapes for each letter. 715 samples, unbalanced. -
lstm.ipynb: A training notebook for the model. -
mouse.py: Python script used to create the dataset. -
test.py: Python script to use the model and type letters.