Introducing Our Beef Yield Estimation Browser Extension

Our new Beef Yield Estimation Browser Extension uses AI to analyze beef meat quality directly from live video streams or images viewed in your browser. The extension categorizes the meat into three quality levels: Red (Poor), Amber/Yellow (Medium), and Green (Premium), providing real-time analysis to help users assess the readiness of beef for maximum yield.

Key Project Tasks

1. ONNX Runtime Web Integration

To enable AI inference capabilities within the browser, we integrated ONNX Runtime Web. This allows the extension to process images and video streams in real-time, providing instant feedback on beef meat quality.

2. Loading the AI Model

A pre-trained AI model is used to classify beef yield into three quality categories. The model processes images and video frames to output one of the three categories: Red, Amber, or Green.

3. Frame Extraction from Live Video Streams

We developed two methods for frame extraction:

• URL-Specific Extraction: Focuses on specific websites where beef evaluation is relevant, extracting frames from HTML tags.
• Universal Extraction: Works across various websites, capturing frames from the entire browser tab for general use.

4. Inferencing and Displaying Results

Once frames are captured, the AI model analyzes them, and the results (red, amber/yellow, green) are displayed on the webpage. Users can instantly assess the quality of beef meat directly within their browser.

5. Optional Automated Bidding

We are exploring the potential for automating beef bidding based on quality estimation. This feature is under further research to assess its practicality and value.

Case Study: User-Driven Corrections and Model Updates

Problem:

Occasionally, the AI model misclassifies meat yield quality, affecting the accuracy of recommendations.

Solution:

To address this, users can manually correct misclassified results, which are then submitted to the server for model retraining. This feedback loop helps to improve the model’s accuracy over time, ensuring continuous learning and better predictions.

Benchmark Results:

Since the introduction of user-driven corrections, the accuracy of yield classification has significantly improved, showcasing the power of continuous learning and user input.

Research and Considerations

• Frame Extraction Techniques: We are evaluating both URL-specific and universal methods to determine the best solution for extracting frames from live video streams or images.
• Automated Bidding Feasibility: We are investigating the potential benefits of automating the bidding process based on beef quality estimation.

Next Steps

• Frame Extraction: Further research is required to refine the frame extraction process and evaluate the best approach.
• Automated Bidding: Begin exploring the feasibility of automating the bidding process based on quality estimation.
• Development: Continue the development of the browser extension, starting with ONNX Runtime Web integration and model loading.

Watch the demo in action on our YouTube video to see the extension in action!