Question

A manufacturing company wants to build a system that inspects photos of circuit boards on the assembly line and flags a specific set of proprietary defect types (e.g., cold solder joints, lifted pads, missing components) unique to their products. The ML team has a labeled dataset of about 5,000 annotated defect images but has minimal ML coding experience and wants the fastest path to a high-accuracy custom classifier deployable as an endpoint. Which Google Cloud approach best fits these requirements?

Accepted Answer

Train a custom image classification model with Vertex AI AutoML using the labeled defect images, then deploy it to an endpoint.

Answer

Use the pre-trained Cloud Vision API label detection to classify the defect types directly.

Answer

Write a custom TensorFlow CNN training script and run it on Vertex AI custom training with a GPU pool.

Answer

Use the Vision API's object localization feature and map bounding boxes to defect categories with post-processing rules.