A common struggle in Mandarin class. We want to store our Chinese notes while also eliminating the laborious process of typing Chinese.

What it does

Given a picture of a Chinese character, our neural net model, built with TensorFlow, identifies the handwritten Chinese characters and creates a Unicode output representing the digital text version of the character.

How we built it

Our IDEs were Jupyter Notebook and VSCode. We built our neural net with Tensorflow and Keras, popular machine learning libraries, and Flask to connect our Python code to our website.

Challenges we ran into

The dataset was too huge and ended up using a lot of time to download, split, train, and test. This caused the project to be incomplete, as we only used a small fraction of available characters.

We also had trouble incorporating a clean front end API and ended up using the basics.

Accomplishments that we're proud of

We were able to create a robust neural net model in less than 4-5 hours.

What we learned

We learned to cut out unnecessary data from the set (who really memorizes 57k Chinese characters?). We also need to make sure our idea hasn't been done already (we used 5+ hours on something already done and had to scrap it).

What's next for Siknow

We entend to conduct more accuracy tests and build a cleaner front end.

Built With

Share this project: