AnimeVsCartoon

The main objective is to train an artificial neural network to classify anime and cartoons

Dataset

The dataset is self-mined using Google Chrome's Fatkun Batch Image Downloader Extension. The dataset contains around 6000 images with almost equal number of images of anime and cartoons

Link to dataset https://drive.google.com/open?id=1_JrLyl3v0B1BNJpJvnBsgDGFjjG2ijyi

Model 1

The model used is a convolutional neural network.

Model 2

This model uses the initial pre-trained layers of VGG-19.

Weights of the above model https://drive.google.com/open?id=1IJFR-lrBhj99z0OB37f2mSheVi5g2cXq

Results

Model 1

Has a 96.07% training accuracy and a 86.67% test accuracy.

Model 2

Has a 99% training accuracy and a 90% test accuracy.

Future Scope

Facial features differ a lot in anime and cartoon. This fact can be exploited to build more efficient models.