Can you share the training strategy on ImageNet

[springsprite]

I appreciate your effort of making these experiment results and sharing code. I am quite interesting in the experiment on ImageNet. I try but cannot produce your result.

My training strategy is based on this repo(https://github.com/PaddlePaddle/PaddleClas/blob/release/2.5/ppcls/configs/ImageNet/MobileNetV3/MobileNetV3_large_x1_0.yaml), which can obtain 75.2% of MobileNetv3-large. I just use the same strategy to train sa-MobileNetv3.

I am looking forward to your rely.

Best wishes!

[SajjadAemmi]

@springsprite Thank you for expressing your interest in my work. Can you share your results and your trained weights with me?