Intelligent defect generation

High-quality defect data is what trains and optimises an inspection model, and on a good production line it is exactly what you do not have. Creating samples by hand gives low authenticity. Waiting for the line to produce them is slow. Changing product type resets the collection. Complex and diverse defects make the whole exercise inefficient.

Defect Synthesis uses AI and diffusion model technology. Forward and backward diffusion algorithms simulate defects of varying type, location and shape from a small number of sample images, and the simulated defects closely resemble real ones in appearance and characteristics.

Operationally it is three steps: mark defect samples, place good product images, synthesise. The generated results carry annotation information, so they go straight into model training with no secondary labelling. Users select defect location, quantity and type themselves.