how to train the code(GAL) in Imagenet dataset?

when we reproduce the work in your paper to purning resnet50 in Imagenet, with the lambda 0.5, the parameters (purned 20301/20688), just one  epoch and the performance is poor. can you help us to reproduce the work in your paper. Thank you so much! 