Transfer learning applied to AlexNet upon EMNIST dataset in Matlab
Download "emnist-balanced-train.csv" from https://www.kaggle.com/crawford/emnist#emnist-balanced-train.csv
After downloading the csv file , create a folder named "dataset" . Within this folder create subfolders each with the names of aplhabets in the classes array present in csv2img.m file . If the alphabet is uppercase , the subfolder name is the same but if it is lowercase then the subfolder name resembles "small_" . Remember to use the correct directory address of the dataset folder being created in the codes.
This file helps in creating dataset from the csv file downloaded .
This file should be executed after creating the dataset . This file applies transfer learning and trains the modified network using the downloaded dataset .
- It would take a lot of time to train the neural network , so using powerful a GPU is recommended.
- This network gave an accuracy close to 89%