chasingbob / character-vision

Using deep learning to recognize characters and digits in random scenarios similar to Google Streetview house number identification but more generic

character-vision

Using deep learning to recognize characters and digits in random scenarios similar to Google Streetview house number identification but more generic.

Objective

To use Deep Learning to identify and recognize characters in any random scenario by only using generated data.

Data

The quality of a deep learning model depends to a large extend on the quality of the data set used for training. Quality data sets are hard to come by, and has rightly been called the new GOLD. Being able to augment existing data and to generate new data if it hasn't already will become an important tool. The main objective of this project is to solve a complicated real world problem solely using generated data.

Model

Architecture

Einstein:

Everything should be made as simple as possible, but not simpler.

Adopting Einstein's quote and changing it slightly for deep learning:

Your model should be made as simple as possible, but as deep as needed.

Fine tuning

Training

Over-fitting

Predict