Beomjin Ahn

Hi, thank you for sharing the code, and I have a question about your implementation. I found that the both of the [m_r](https://github.com/mindslab-ai/hififace/blob/c92db89c7bd0e4d1518f872827fdb69a98f16060/model/hififace.py#L154) and [m_low](https://github.com/mindslab-ai/hififace/blob/c92db89c7bd0e4d1518f872827fdb69a98f16060/model/hififace.py#L169) are generated by tanh function...

Keep-rate scheduling of DropBlock in a multi-GPU environment

2

Hello, I found an issue while trying to train your model. In your code, the variable ['self.num_batches_tracked'](https://github.com/kjunelee/MetaOptNet/blob/7a8e2ae25ef47cfe75a6fe8bc7920dc9fd29191f/models/ResNet12_embedding.py#L32) should count the progress of the episode by [increasing](https://github.com/kjunelee/MetaOptNet/blob/7a8e2ae25ef47cfe75a6fe8bc7920dc9fd29191f/models/ResNet12_embedding.py#L38) when the model is...