Table of Contents

• 1. Setting up the project dependencies
• 2. Running the code
  • 2.1. Using the jupyter notebook
  • 2.2. Running the code using CLI
• 3. Results
  • 3.1. Model Architecture
  • 3.2. Samples
  • 3.3. Accuracy
• 4. TODO
• 5. Author

Vowel consonant classification

This repository contains code and results for detecting vowels and consonants in Tamil language characters.

Diving into the code

1. Setting up the project dependencies

1. Install pipenv using the command pip insalll pipenv
2. Install the project dependencies by running the command pipenv install in the project directory which will set up the dependencies after creating a virtual env.

2. Running the code

2.1. Using the jupyter notebook

You can find the jupyter notebook for this project in notebooks/Vowel_consonant_classification.ipynb. Running it should be fairly simple as the code is well-commented and explained fairly well.

2.2. Running the code using CLI

In the project directory use the following commands:

For training: vc train

For testing: vc eval <model_path> <test_dataset_path>

<model_path> is the path to the saved pytorch model. One such model is in the classifier/results/ directory

<test_dataset_path> is the path to a directory containing the test images.

The file config.yaml stores the configuration of the classifier.

3. Results

3.1. Model Architecture

Used the InceptionV3 network for fine-tuning on the dataset hosted at https://www.kaggle.com/c/padhai-tamil-vowel-consonant-classification/data

The architecture is shown below:

where X is the input image, X1 are the extracted features which are passed through two classifiers for detecting the Vowel and the consonants. The number of Vowels in Tamil language are:13 and the consonant count is 19.

3.2. Samples

Some samples from the training dataset are shown in the figure below. The training data used is a synthetic dataset.

3.3. Accuracy

The test set accuracy on the dataset from this method is around 94.04% on a test set size of around 10000 samples.

4. TODO

1. Compute the accuracy after introducing Mixup data augmentation and compute the results.
2. Try out other model architectures like ResNeXT, WideResNet and ResNet-152 and compare the results

5. Author

Kushagra Pandey / @kpandey008