Domain : Computer Vision, Machine Learning Sub-Domain : Deep Learning, Image Recognition Techniques : Deep Convolutional Neural Network Application : Image Recognition, Image Classification
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.
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 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 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 |
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 |
Dataset | Training | Validation | Test |
---|---|---|---|
Accuracy | 99.71% | 98.74% | 93.72% |
Loss | 0.0234 | 0.0219 | --- |
Precision | --- | --- | --- |
Recall | --- | --- | --- |
Roc-Auc | --- | --- | --- |
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 |
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 |
![](https://github.com/anjanatiha/Kannada-MNIST-Classification-with-Deep-Learning/raw/master/demo/images/class-dist.png)
![](https://github.com/anjanatiha/Kannada-MNIST-Classification-with-Deep-Learning/raw/master/demo/report/hist.png)
Languages : Python Tools/IDE : Kaggle Libraries : Keras
Duration : February 2020 Current Version : v1.0.0.10 Last Update : 02.12.2020