In these notebooks we will learn how to implement a convolutional neural network (CNN) regressor to localize digits of the MNIST dataset. We will use the PyTorch library for training our model.
- The input to our model will be a 64 x 64 image with a MNIST digit at any location, and the output of the model are four real numbers that define a bounding box (x, y, width, and height) around the digit.
- Then we will modify our model by adding a classification output, so that it can jointly predict the bounding box of the digit and also its class label.
The above notebooks are adapted versions of an assignment given by Lluis Gomez i Bigorda for my M5:Visual Recognition class