Skip to content

ml5 feature-extractor POC for icon search prediction based on canvas draw.

License

Notifications You must be signed in to change notification settings

en3sis/ml-classificator

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

About

This is a POC using Machine Learning ml5 and MobilNet model with the idea to train the model and predict labels based on users input.

Live

You can try the live version at https://ml-classificator.vercel.app/

Run locally:

First, make sure you install the dependencies:

npm i

Once you do that, you can now run the local server and serve the static files by running:

npm start

Current state:

At this moment, we have a dataset of 4 categories, arrow-left, arrow-right, clock, x. Once the project loads, click on the Train button (open the devtool so you can see the feedback in the console)

Once the output is Loss is null you can now draw an arrow (left or right) and click on predict. You can cick on Reset and draw a new shape.

TODO:

  • Add more feedback when the models is ready to train or it finishes commit
  • Add a more dynamic way to load the dataset for training commit.
  • Load pre-trained model.
  • Add field that allows you to rename the currect canvas image.