The suitable lr for mobilenet_v2 and efficientNet

[detectRecog]

Hello, thank you for your amazing work! In practice, I find that the lr for mobilenet_v1 in the demo training command is not suitable for mobilenet_v2 and efficientNet, resulting in unstable training. Could you provide your lr choices for other backbones?

Thanks in advance.

[haihabi]

Hi, thanks for your comments.

The hyperparameters are shown in the paper https://www.ecva.net/papers/eccv_2020/papers_ECCV/html/5414_ECCV_2020_paper.php

Is that helpful?