Scene Generation Pipeline Overview

Goals

Automate the generation of simulation-ready .json files by:

• Segmenting satellite images
• Deciding object placements using rules
• Outputting properly aligned Vega Prime scene configs

---

Current Progress (as of May 18, 2025)

• Extracted the largest mask from the SAM2-based segmentation output of a satellite image.
• Intended next step: Automate the generation of a JSON prompt for object placement using the GPT API.
• Issue encountered: Unable to configure the correct libraries and imports to make API calls locally.
• Temporary workaround:
• Manually uploaded the 2D mask corresponding to the largest contour.
• Labeled it and plan to input it manually into GPT to receive a JSON prompt.
• Next steps:
• Resolve API configuration issues.
• Automate the process end-to-end.
• Apply the pipeline to 4–5 more images to test consistency and scalability.