batch-picture-to-md

Batch image layout analysis is automated by processing research figure images from pdf-ocr artifacts to infer layouts, perform visual auditing, and generate per-image analysis reports. The workflow groups images into batches and delegates long tasks to a generalist agent.

Multiple pdf-ocr image artifacts are converted into a markdown report by batching 3-5 images per group, generating safe JSON for per-image layout analysis, and merging the results into a final report.

Yes, batch processing for research figure images supports an optional multi-agent orchestration workflow. It delegates long tasks to a generalist agent and groups images to optimize token usage during analysis.

Automated layout inference works with image artifacts, specifically targeting files like image_*.png within the artifacts/ directory. It performs visual layout determination and self-calibration to ensure bounding boxes cover content.

The visual audit requires safe JSON generation to ensure structured per-image analysis and prevent parsing errors before merging. This validates automated layout inference and annotation generation into a final report.

Batch grouping for image layout determination limits processing to 3-5 images per batch to optimize token usage. Long tasks require delegation to a generalist agent, which may increase overall processing time.