Xa9aX ツ

https://github.com/BangguWu/ECANet/issues/46 @konioy @ndsclark

Based on my experience of loading the weights for ECANet-50, it seems the authors have set up the kernel sizes manually based on the input tensor size to the bottleneck...

@tycallen please refer to this - https://github.com/ultralytics/yolov3/issues/1098#issuecomment-620149884

Sure, yes

It seems that the `copy.deepcopy(list(strategy.model.parameters()))` does work but returns the params as a list cloned from the source generator object in the `evaluation/metrics/checkpoint.py` which when used in `logging/wandb_logger.py` doesn't seem...

@AndreaCossu I am no longer at W&B. But I am happy to finish this but won't be able to before December. Thanks!

@anhle-uet The performance in terms of generalization capability of the network is mirrored for both the normal implementation and the CUDA implementation, however, the CUDA implementation is much faster.

Additionally considering GPT-3 size I would suggest not to use any optimizer above SGD because of the computation levels. Same goes for Mish.

@LifeIsStrange everything above SGD is expensive.

@hellock any updates on this?