Deep_Doc_Classifier

Dataset setup

To train your network, you need to first download the Tobacco-3482 dataset with 10 document classes. You can use the train_test_split.py to split the dataset into training and testing folders. Give the path to the Dataset folder as shown in line 7 of train_test_split.py

dataset_path = Path(r'Data/Tobacco3482-jpg')

Requirements:

• tensorflow
• keras
• pytorch
• torchvision
• pandas
• matplotlib
• numpy
• PIL
• seaborn

Files Description:

data_modification.py

The file data_modification changes the dataset images from a single chanel to 3 channel. This is done as an input to alex net. Also the image size is reduced to 227x227 inorder to reduce the computational overhead.

config.py

The file config is used to initialize hyperparameters along with other parameters, such as number of epochs, batch-size, and the resize value for the image.

train_test.py

The train_test file is used to train the model on the new dataset.

DeepDoc_AlexNet.ipynb

This is the main execution file for the deep doc classifier using the pretrained alex net (pretrained on Imagenet), for classifying documents. There are total of 10 classes.