Improving surgical phase recognition using self-supervised deep learning.

In contrast, BYOL, by design, does not rely on large batch sizes for feature learning, so the batch size configuration was not as significant.

also in 271 other papers