Distinguishing artefacts: evaluating the saturation point of convolutional neural networks

Prior work has shown Convolutional Neural Networks (CNNs) trained on surrogate Computer Aided Design (CAD) models are able to detect and classify real-world artefacts from photographs. The applications of which support twinning digital physical assets in design, including rapid extraction part geometry model repositories, information search \& retrieval identifying components the field for maintenance, repair, recording. performance CNNs classification tasks have been dependent training data set size number classes. Where prior works used relatively small sets ($<100$ models), question remains as ability a CNN differentiate between increasingly large repositories. This paper presents method generating synthetic image online CAD further investigates capacity an off-the-shelf architecture class increases. 1,000 were curated processed generate scale sets, featuring coverage at steps 10$^{\circ}$, 30$^{\circ}$, 60$^{\circ}$, 120$^{\circ}$ degrees. findings demonstrate capability computer vision algorithms repositories up 200, beyond this point CNN's is observed deteriorate significantly, limiting its present automated artefacts. Although, match more often found top-5 results showing potential models.