A FastAI ResNet-18 implemented and trained on Steel Defect Database by NEU The Datasets is from Kechen Song, Northeastern University (NEU).
You can visit their Homepage to download 1800 images.
FasAI is very powerfull and easy to use library built on top of PyTorch. I have trained a ResNet-18 on this database. The results and further information about code is in the notebook and the database and further information about it, can be donwloaded from the link provided.
Note: This code only wors with FastAI version 1. I will try to update it to work with FastAI version 2.