This repository trains a LSTM Model for the Skribbl.ai web app.

The data used comes from the Quick, Draw! dataset provided by Google.

Raw data is given in stroke format:

[
  [  // First stroke
    [x0, x1, x2, x3, ...],
    [y0, y1, y2, y3, ...],
    [t0, t1, t2, t3, ...]
  ],
  [  // Second stroke
    [x0, x1, x2, x3, ...],
    [y0, y1, y2, y3, ...],
    [t0, t1, t2, t3, ...]
  ],
  ... // Additional strokes
]

We want to perform normalization on the stroke data using a variant of the RDP algorithm to construct a stroke format compatible with the LSTM model:

[ // First stroke [x0_norm, y0_norm, is_end_1], [x1_norm, y2_norm, is_end_2], [x0_norm, y0_norm, is_end_3], ........ ]......

We then pad each stroke using the post-padding technique, to ensure consistency of stroke size for LSTM model.