Skip to content

cs-util-com/VisionValidator

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

93 Commits
 
 
 
 
 
 
 
 

Repository files navigation

VisionValidator

VisionValidator is a single-page web application that evaluates and compares how different AI vision models perform object detection tasks by returning bounding box data.

Overview

This tool helps developers, researchers and AI enthusiasts:

  • Test various AI vision models with your own images
  • Validate the returned bounding box data against a schema
  • Visualize object detection results with bounding boxes
  • Compare consistency across multiple queries with heat maps
  • Analyze contextual attributes identified in the image

Features

  • Multi-Model Support: Works with OpenAI GPT-4o, Google Gemini, and Anthropic Claude 3
  • Bounding Box Visualization: Displays detected objects with colored bounding boxes
  • Heat Map Generation: Creates visual representations of AI consistency across multiple queries
  • Schema Validation: Ensures AI responses conform to a standardized format
  • Secure API Key Management: Stores API keys locally in your browser
  • Multiple Query Analysis: Makes 5 identical requests to analyze model consistency
  • Response Comparison: Toggle between different AI responses for the same image

Getting Started

Usage

  1. Select an AI Model: Choose between OpenAI GPT-4o, Google Gemini, or Anthropic Claude 3
  2. Enter Your API Key: Provide your API key for the selected service (stored locally in your browser)
  3. Upload an Image: Select any image you wish to analyze
  4. Process the Image: Click "Send to AI & Validate" to start analysis
  5. View Results:
    • See bounding boxes drawn on your image
    • View the heat map showing detection consistency
    • Toggle between different AI responses using the selector
    • Review the full JSON response below

Requirements

Technical Details

VisionValidator is a frontend-only application built with:

  • HTML, JavaScript, and Tailwind CSS
  • LangChain.js for AI model integration
  • Ajv for JSON schema validation
  • Canvas API for visualization

Privacy Considerations

  • All processing happens in your browser
  • Images never touch our servers - they're sent directly from your browser to the AI provider
  • API keys are stored in your browser's localStorage and only sent to their respective providers

Examples

Use Cases

  • AI Research: Evaluate and compare different vision models
  • Quality Assurance: Test consistency of object detection across multiple queries
  • Model Evaluation: Determine which AI performs best for specific image types
  • Educational Tool: Learn about how AI systems interpret visual information

Contributing

Contributions are welcome! Feel free to submit issues or pull requests.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • HTML 100.0%