This my solution to the modified MNIST project on Kaggle(https://www.kaggle.com/c/modified-mnist) The goal is to create classifiers to analyze images consisting of two digits and a character. The character may be an 'a' or 'A' (for add), in which case your classifier has to return the sum of the two digits, or an 'm' or 'M' (for multiply), in which case your classifier has to return the product of the two digits. Additionally, a variety of different textures are used in the backgrounds of the images.
To run this solution 1- create a tensorflow-gpu python environemnt. 2-Download the data from https://www.kaggle.com/c/comp551-modified-mnist/data. 3- Run ConvertCSVToTFRecord.py 4- Run ModifiedMNIST_CNN
Good luck