Bedrock : analyse graph image and answer questions about it

Yes, Amazon Bedrock offers several foundation models (FMs) with multimodal capabilities that can analyze graph images and answer questions about them. Specifically, Anthropic's Claude 3 Sonnet and Meta Llama 3.2 90B Vision model have demonstrated impressive capabilities in processing and interpreting visual content like graphs, diagrams, and tables.

These models can handle your use case effectively:

• You can input a picture containing a simple graph
• Provide additional context or details about the graph content
• Ask questions about the graph, including theoretical aspects of the graph type or business interpretations of the data

For example, you could upload a population distribution graph and ask questions like "Which city has the highest number in a specific range?" or "Compare the demographics between different regions." The models can extract contextual information from the graph, even when explicit labels might be absent, and provide insights through natural language responses.

These capabilities are particularly useful for:

• Visual question answering
• Chart and graph interpretation
• Extracting structured information from visual data
• Combining visual information with additional context to provide comprehensive analysis

To get started with these capabilities, you can use the chat playground feature on Amazon Bedrock, which allows you to interact with and extract information from images directly.
Sources
Read graphs, diagrams, tables, and scanned pages using multimodal prompts in Amazon Bedrock | AWS Machine Learning Blog
Context extraction from image files in Amazon Q Business using LLMs | Artificial Intelligence
Best practices for Meta Llama 3.2 multimodal fine-tuning on Amazon Bedrock | AWS Machine Learning Blog