By using sklearn/tensorflow/pytorch libraries, we generated a classifier that could classify hand-drawn images from Google’s Quick, Draw! project. The dataseet contained images that were not always centerred and straight, or free of noisy artifacts. In addition, the classifierrr needed to identify empty pictures as well.
Our implementation of classifier built on a deep learning framework improved accuracy by 11% above a baseline CNN classifier. We utilized several algorithms, such as CNNs and played around with utilizing pre-trained models such as Google Inception V3.