(Day 33) Tensorflow deployment specialization and a webapp for recognizing the Korean alphabet

Ivan Ivanov · February 3, 2024

Hello :) Today is Day 33!

A quick summary of today:

Firstly - learning how to deploy tensorflow models

image

I learned how to train a machine learning model in the browser. I can do it right away with tensorflowjs.

You can train it directly in the browser

image

image

You can load a model developed in a jupyter notebook too.

image

One of the assignments was to create a rock-paper-scissors model using my laptop’s camera.

image

image

image

(no idea why balloons appeared on that last one)

Secondly, deployed a simple Hangul classification model

I found out if I was going up the model with js today, so I wondered if I could actually come up with an idea I’ve made before. Hangul classification model < I made it when I first learned machine learning (Summer 2023~) and I wanted to make a simple Hangul character classification app or website based on this model even if I didn’t understand 100% what I did at that time. But I didn’t have much time, so I gave up. Today, this idea was born again.

After constructing a simple model, I made a website with html and js The accuracy and loss of the model are as follows

image

I had to deal with some problem with canvas from my browser which was frustrating, but I finally managed to get it work.

image

The accuracy and loss of the model were not so good, so I tested it with the characters I wrote.

image

It got 20 out of 30. Not too bad haha.


That is all for today!

See you tomorrow :)

Original post in Korean