The model used in the Digit Recognition Web App can recognize which numbers were drawn by you. It was trained using two python frameworks for Machine Learning and Deep Learning Tensorflow, Keras and another JavaScript framework used for inference the TensorflowJS.
Anaxímeno Brito
36 x 36 Image
For each drawing, the model outputs ten probabilities corresponding to the chance of being each of the ten digits (0 to 9), and the digit with greater probability will be considered as the result.
September 2021
Mnet V3
Convolutional Neural Network based architecture
The Digit Recognition Model was built to be used on this web application and explore the application of Deep Learning in real-life situations. The intended users are persons who want to test this Web App or explore this model's capabilities.
The data used to train was the MNIST dataset, which consists of a set of black and white images with digits between 0 to 9. Since the digits are centered on the image, it was necessary to apply certain arbitrary transformations before train the model like translations, zooms, and rotations, which made the model predict quite well on different conditions. I also applied another transformation during training, increasing images sizes to 36x36 because I believe it is a better input format for this application.
If you want to see the how it was trained just click this link.