Flickering is caused by the browser's screenshot function and does not affect the software's operation.

VisionDeck

An autonomous Poker Bot (more specifically for the Deuces Wild variant) powered by Computer Vision and AI.

Note: Warning: This software was developed as a proof of concept for educational purposes. Do not try to use it in real situations or you will end up in trouble.

Key Features • Setup • Credits • Donate • License

Key Features

• Autonomous operation powered by Playwright

• Card identification algorithm using OpenCV

• 2 operation modes: Internal logic (roughly based on Not So Ugly Ducks Strategy by Wizard of Odds; may have some flaws as I'm not a poker expert) and AI solving (OpenAI or Anthropic)

Setup

Prerequisites

Make sure you have a working installation of Git and Node.js. You will also need the API key of an AI provider if you want to use AI mode.

Environment Variables

Variable	Description
AI_MODE	Whether to use an AI provider to solve the poker hand. Defaults to false.
AI_PROVIDER	Which AI provider to use if using AI mode. Possible values are openai and anthropic.
ANTHROPIC_API_KEY	API key to use if the provider is Anthropic.
OPENAI_API_KEY	API key to use if the provider is OpenAI.

Step by step

• From your command line:

# Clone this repository
$ git clone https://github.com/Leandro-Amorim/visiondeck

# Go into the repository
$ cd visiondeck

# Install dependencies
$ npm install

# Run the app
$ npm run dev

• Follow the instructions listed on the screen (keep the canvas fully on the screen, select the "Deuces Wild" variant and wait for the "Press Deal to Start" message to appear).

• If you ever need to regenerate the training data, you can do so by adding the card images to the unprocessed-cards folder and running the command:

# Regenerate training data
$ npm run process

Make sure you follow the naming of the files already in the folder.

Credits

• 'Not So Ugly Ducks' Deuces Wild Strategy by Wizard of Odds
• OpenCV-Playing-Card-Detector
• OpenCV
• Playwright
• Node.js

Donate

If you think this project has helped you in any way or you've learned something new, consider buying me a coffee, I love it!

License

Distributed under the MIT License. See LICENSE.txt for more information.