Identify characters from Google Street View images
Competition focuses on the task of identifying characters from Google Street View images. It differs from traditional character recognition because the data set contains different character fonts and the background is not the same for all images.
With the below methods I have held a rank of [9/53] among the top 17%
The purpose of the competition was to identify characters from Google Street View images. The data was taken from the Chars74K dataset, which consists of images of characters selected from Google Street View images.
- Performed Data visualization and transformation to remove skewness and correlation.
- One Hot Encoding was performed on categorical data to convert it into a binary vector to be used as a feature in the Machine Learning Model.
- Implemented a CNN based model to tackle the problem, resulting in 82.04% training and 81.58% validation accuracy. Model achieved 77.97% accuracy on the test dataset.
- Final model submitted was a CNN trained on augmented image data along with hyper tuning of parameter resulting in 87.74% training and 84.73% validation accuracy. Model achieved 82.98 % accuracy on the test dataset.
Exploratory data analysis :
Code repo :