Anime vs Cartoon

Inspiration

My friends who do not watch Anime consider them as Cartoons. "Is Anime the same as Cartoon?" is always a topic of debate among us. They are simply unable to distinguish between the two. To solve this problem, I've created this Project.

What it does

The final project takes an Image as an Input and Classifies it as Anime or Cartoon and outputs the Probability score for each class. It has a simple UI and someone with basic knowledge of computers can easily operate it.
Try it Yourself - https://animevscartoon.herokuapp.com/

How we built it

First, we collected images of the most popular Anime and Cartoon shows by scrapping them from Google Images.
The next task was to build the Deep Learning Model which consumed my entire Saturday.
Then on Sunday, we worked on the UI which is built using HTML, CSS, and JS, and Flask is used at the backend. In the end, we deployed our app on Heroku and created a video presentation for it.

Challenges we ran into

The first difficulty we faced was choosing the hyperparameters for the model. I have literally trained it like 20-25 times and then arrived at the optimal parameters that essentially forms the final model.
The next difficulty was finding a way to pass images to the model on the server. On the system it's easy but on the cloud, the normal way is to use a Cloud Storage Bucket but Unfortunately, in the APAC region, the Google cloud offer for $100 was not valid and we did not have any credits. Eventually, we passed the image in base64 format and converted it back to an OpenCV image object with a few lines of python code.

Accomplishments that we're proud of

Although our model's overall Accuracy is 87% on all the images we personally tried it on (around 100 images), it is yet to produce any incorrect results, which is pretty awesome.

What we learned

I learned how changing the learning rate of an inbuilt optimizer can affect the performance of the model so much.
The base64 method of passing an image to a model is a discovery to cherish forever.