SPACEtomo training dataset for lamella detection using YOLOv8

Training data used to train a YOLOv8 model for cryoFIB-milled lamella detection on whole grid TEM montages.

This dataset contains 614 LM map pieces of varying pixel sizes (rescaled to 400 nm/pixel) containing 1076 examples of lamellae. Lamellae were classified into "good", "contaminated", "thick" and "broken" lamellae. An additional 534 LM map pieces not containing any lamellae were added to the training dataset. To enhance rotational invariance, the 614 LM map pieces were added to the training dataset again with flipped axes.

Thanks to Matthias Pöge, Gregor Weiss, Sven Klumpe, Anna Bieber and Cristina Capitanio for providing whole grid TEM maps.