physical-ai-defect-image-generation

Automate defect-image generation pipelines for PCB AOI datasets by orchestrating end-to-end workflows with NVIDIA OSMO, coordinating data ingestion, rendering, augmentation, and labeling deterministically across multiple flows.

Day 0 flows in AI defect-image generation cover texture defects, good images, and structural defects, while Day 1 flows handle real-photo alignment and manual ROI operations for refined anomaly labeling.

OSMO defect-image orchestration supports PCBA, metal surface, and glass use cases, applying unified pipeline execution and artifact management across these distinct anomaly generation workflows.

Submit the Day 0 texture_defect_generation.yaml file with required parameters including dig_url_root, board, image_edit_endpoint, and anomaly_types_json to start the end-to-end OSMO workflow.

Running an OSMO anomaly generation pipeline requires preflight checks for credentials, pod template, and URL artifacts, along with memory stamping and explicit osmo submit knobs for deterministic outputs.

Coordinate multiple components including usd2roi, image-edit, and anomalygen alongside per-board cookbooks in a single defect-image pipeline to enable scalable execution and organized output under a designated OSMO root.