tao-run-deft-aoi

You can automate the PCB inspection model improvement loop by running the DEFT AOI process, which orchestrates baseline evaluation, root cause analysis, synthetic defect generation, retraining, and deployment using NVIDIA TAO VisualChangeNet models.

The DEFT AOI improvement loop is an automated process for PCB inspection models that iteratively performs baseline evaluation, root cause analysis, synthetic defect generation via AnomalyGen, data mining, retraining, and deployment until KPI targets are met.

Yes, Docker and nvidia-container-toolkit are required dependencies to run the DEFT AOI automation loop, along with tao-toolkit-pyt, tao-toolkit-deploy, and tao-toolkit-ds for executing specific pipeline stages.

You can set and adjust custom KPI targets for false-accept rate and recall within the DEFT AOI improvement loop, allowing the automated retraining process to fine-tune PCB inspection models until your specific metrics are satisfied.

Yes, the DEFT AOI improvement loop integrates with HuggingFace to manage model weights for NVIDIA TAO VisualChangeNet and ChangeNet PCB inspection models during the automated retraining and deployment pipeline.

The DEFT AOI loop uses AnomalyGen to generate synthetic defects during the automated improvement process, combining synthetic data generation with data mining to retrain PCB inspection models for better accuracy.