physical-ai-defect-image-generation

What problem does it solve?

Orchestrates end-to-end defect image generation pipelines for AOI datasets on NVIDIA OSMO, coordinating USD asset management, dataset prep, image-edit augmentation, anomaly generation, and labeling to produce ready-to-use anomaly trees and training data.

Core Features & Use Cases

• Supports Day 0 texture defects, Day 0 good-image, Day 0 structural defects, and Day 1 real alignment or manual ROI flows for PCBA, plus metal surface and glass use cases.
• Handles per-board cookbooks, pretrained vs finetune modes, and outputs organized under the DIG URL root for reproducibility and traceability.
• Provides preflight gates, memory rules, and monitoring hooks to ensure safe, repeatable runs in production.

Quick Start

Run a Day 0 texture defect flow using the shipped PCB dataset and in-cluster Image-Edit endpoint to produce the first anomaly-inference run.