AI-102 Question 466
Select 2You are working on a document processing solution for a client using Azure OpenAI. The client requires the solution to process and analyze scanned invoices containing text and images, extracting key information such as item descriptions, prices, and totals. Which of the following steps should you take to implement the solution using a large multimodal model in Azure OpenAI?
- A
Use the Azure OpenAI GPT-4 model with visual capabilities to process both text and images from the invoices.
- B
Preprocess the scanned invoices by converting them into text-only formats before using Azure OpenAI.
- C
Integrate Azure Cognitive Services Computer Vision to extract text from the images, then use Azure OpenAI for further analysis.
- D
Use the Azure OpenAI DALL-E model to generate synthetic invoice images for training your solution.
- E
Leverage the Azure OpenAI ChatGPT model to parse and interpret the extracted text and context of the invoices.
Show answer and explanation
Correct answers: A, C
Explanation
To process and analyze scanned invoices containing both text and images, the solution must leverage multimodal capabilities. Azure OpenAI GPT-4 with visual capabilities can directly process multimodal inputs, while Azure Cognitive Services Computer Vision can be used to pre-extract text from images if needed. Other models like DALL-E and ChatGPT are not designed for this type of task, and preprocessing by converting to text-only formats would discard valuable visual context.
- A. Correct.
The Azure OpenAI GPT-4 model with visual capabilities supports multimodal inputs, meaning it can process both text and images, making it suitable for analyzing scanned invoices.
- B. Incorrect.
Converting scanned invoices into text-only formats before using Azure OpenAI limits the ability to process visual elements and does not leverage the multimodal capabilities of large models.
- C. Correct.
Integrating Azure Cognitive Services Computer Vision to extract text from images is a valid step to preprocess image data before further analysis with Azure OpenAI.
- D. Incorrect.
The DALL-E model is designed for generating images based on text prompts; it is not relevant for extracting or analyzing information from scanned invoices.
- E. Incorrect.
The ChatGPT model is designed for conversational AI and cannot process visual inputs such as scanned invoices, making it unsuitable for this scenario.