Question : super-convergence

[pfeatherstone]

Hi @AlexeyAB , have you seen the papers (https://arxiv.org/pdf/1708.07120.pdf) and implementations (https://pytorch.org/docs/stable/_modules/torch/optim/lr_scheduler.html#OneCycleLR) of super-convergence? I'm really interested in training yolov3 in less than 100 epochs. Typically, it requires almost 300 epochs which is painful. Is this something you intend on implementing in darknet ?

[AlexeyAB]

• paper: https://arxiv.org/abs/1708.07120v3
• source: https://github.com/lnsmith54/super-convergence
• small description: https://towardsdatascience.com/https-medium-com-super-convergence-very-fast-training-of-neural-networks-using-large-learning-rates-decb689b9eb0