Closed JiaxiongQ closed 4 years ago
How did you train Ghost-ResNet-50 (s=2) on Imagenet to achieve 75.0% top-1 accuracy? Did you use cosine learning rate and label smooth?
We used multi-step learning rate without label smooth.
How did you train Ghost-ResNet-50 (s=2) on Imagenet to achieve 75.0% top-1 accuracy? Did you use cosine learning rate and label smooth?