Kannada MNIST Classification with Deep Learning (Custom Convolutional Neural Network)

Domain             : Computer Vision, Machine Learning
Sub-Domain         : Deep Learning, Image Recognition
Techniques         : Deep Convolutional Neural Network
Application        : Image Recognition, Image Classification

Description

Kannada is a language spoken predominantly by people of Karnataka in southwestern India. The language has roughly 45 million native speakers and is written using the Kannada script.
1. Detected 10 Kannada (Kannada is a language spoken predominantly by people of Karnataka in southwestern India. The language has roughly 45 million native speakers) digits from images with Custom Convolutional Neural Network.
2. Built a simple custom convolutional neural network with few 2D convolutional, Maxpool and 1 dense layer with around 888K trainable params.
3. After 27 training iterations, attained testing accuracy of 97.70% and loss 0.03 on 60K (12MB+) OCR image dataset.

Code

GitHub Link      : Kannada MNIST Classification with Deep Learning (GitHub)
GitLab Link      : Kannada MNIST Classification with Deep Learning (GitLab)
Kaggle Notebook  : Kannada MNIST Classification with Deep Learning
Portfolio        : Anjana Tiha's Portfolio

Dataset

Dataset Name     : Kannada MNIST
Dataset Link     : Kannada-MNIST (Kaggle)
                 : 
                 
Original Paper   : Kannada-MNIST: A new handwritten digits dataset for the Kannada language 
                   Authors: Vinay Uday Prabhu 

Dataset Details

Dataset Name            : Kannada MNIST
Number of Class         : 10

Dataset Subtype	Number of Image	Size of Images (GB/Gigabyte)
Total	40,000	12 MB
Training	34,000	10.2 MB
Validation	6,000	1.8 MB
Testing	44,004

Model and Training Prameters

Current Parameters	Value
Base Model	Custom CNN
Optimizers	Adam
Loss Function	Categorical Crossentropy
Learning Rate	0.0001
Batch Size	128
Number of Epochs	27
Training Time	9 min

Model Performance Metrics (Prediction/ Recognition / Classification)

Dataset	Training	Validation	Test
Accuracy	99.71%	98.74%	93.72%
Loss	0.0234	0.0219	---
Precision	---	---	---
Recall	---	---	---
Roc-Auc	---	---	---

Other Experimented Model and Training Prameters

Parameters (Experimented)	Value
Base Models	Custom Convolutional Neural Network wwith 888K params
Optimizers	Adam
Loss Function	Categorical Crossentropy
Learning Rate	0.01, 0.001, 0.0001
Batch Size	32, 64, 96, 128, 256
Number of Epochs	27 - 100
Training Time	9min

Hardware

Parameters (Experimented)	Value
Platform	Cloud/Online
Platform Name	Kaggle Notebook
GPU Brand	NVidea
Model Name	Tesla P100-PCIE-16GB
Memory	16 GB
Number of Core	2

Visualization

Class Distribution:

Model Performance:

Tools / Libraries

Languages               : Python
Tools/IDE               : Kaggle
Libraries               : Keras

Dates

Duration                : February 2020 
Current Version         : v1.0.0.10
Last Update             : 02.12.2020