In most manufacturing facilities, a number of defects get examined manually by QA engineers.

Some types of defects have a non-representative amount of samples, high-speed production lines are challenging for visual inspection by a human, traditional sensors can only detect standardized defects, items on the line may have a variety of package designs or form - all of this makes human inspection required. However, manual inspections are not perfect either: decisions of different engineers about the same product do not always coincide. Industrial quality inspection is particularly suitable for AI due to its repetitive nature and high level of predictability. NVIDIA Metropolis enables data labeling, CV models training, models integration with the engineer’s workplace, and automatic decision making. Self-supervised neural networks can learn directly from the production line video stream, and can be trained to detect anomalies at high speed and without human labeling.