Scribbl

Gallery

About Scribbl​

Technical Details​

Scribbl uses a pre-trained TensorFlow Lite model to run inside the Cartesi Machine allowing good performance in a constrained environment. The model is trained over 100 categories of the infamous Quick Draw dataset open-sourced by Google.

Quick Draw Dataset​

It is a collection of 50 million drawings across 345 categories, contributed by players of the game Quick, Draw!. The drawings were captured as timestamped vectors, tagged with metadata including what the player was asked to draw and in which country the player was located. You can browse the recognized drawings on quickdraw.withgoogle.com/data.

Python Backend​

We use Pillow and NumPy libraries to pre-process images received at the backend machine. The processed image in a 28x28 resolution is then fed to the inference to classify and output its predictions. Top 3 predictions are submitted as an ABI encoded data to the smart contract onchain.

Scribbl Smart Contract​

We define a contract called ScribbleTaskManager that relays inputs to our coprocessor backend and also emits the output events. It also calculates the running average score of the user. Its public functions are available to the frontend client to fetch the results and leaderboard scores.

NextJS Frontend​

The frontend webapp available at scribbl.fun is written in Typescript with Wagmi library for web3 interactions. It uses Hero UI as a CSS framework.

What's next​

We want to make Scribbl the simplest and the most fun app out there. We that in mind we want to focus on user experience as priority which includes showcase of user data, better accuracy, more categories and login abstractions. Other simple features that are on our mind include time based challenge and minting your doodles as NFTs.

We're open for suggestions, reach out to us. Let's make eet happen!