This repository provides altered roman numeric dataset and code used during the Data-Centric AI Competition in Python. I improved model accuracy from the baseline of 0.64421 to 0.81240, placing at ranking 103/516.
There are 3 main steps involved in generating this dataset.
- Manually correcting mislabeled and removing inappropriate samples from the original data
- Applying data augmentation techniques by using Augmentor. This work employs ...
- Skimming through the dataset and removing improper samples again