Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/hasibzunair/unconventional-wisdom

[ICBSLP'2018] 6th place solution to Kaggle Bengali Handwritten Higit Recognition
https://github.com/hasibzunair/unconventional-wisdom

computer-vision deep-learning neural-network transfer-learning

Last synced: about 2 months ago
JSON representation

[ICBSLP'2018] 6th place solution to Kaggle Bengali Handwritten Higit Recognition

Awesome Lists containing this project

README

        

## What is it about?

Sixth place solution to kaggle bengali handwritten digit recognition. Achieved 97.606%. A VGG16 architecture pre-trained on imagenet used as a baseline with hyperparameter tuning, fine tuning intermediate layers, data augmentation and test time augmentation.

The unconventional approach led to surprising results which caught Jeremy Howard's attention which he ended up [tweeting about](https://twitter.com/jeremyphoward/status/1050427625011703808)!

## Usage

This architecture is implemented in Python 3.6 and Keras using Tensorflow as backend.

### Dependencies

Tested code using:

* Ubuntu 14.04
* Python 3.6

### Directory Structure & Usage
* `codes`: Contains codes to final submission
* `others`: Contains helper codes and experimental notebooks. VERY MESSY!

## Data set
* [Numta DB](https://www.kaggle.com/c/numta/data)

## Cite

If you find this work useful in your research, please consider citing:
```
@inproceedings{zunair2018unconventional,
title={Unconventional Wisdom: A New Transfer Learning Approach Applied to Bengali Numeral Classification},
author={Zunair, Hasib and Mohammed, Nabeel and Momen, Sifat},
booktitle={2018 International Conference on Bangla Speech and Language Processing (ICBSLP)},
pages={1--6},
year={2018},
organization={IEEE}
}
```
You can also find it in IEEE Xplore Digital Library [here](https://ieeexplore.ieee.org/document/8554435)