The training results can not reach the author's effect #89
Description
Why can't I get the same result as the author on ade20k data set by using the split code? The result I get is only more than 30, which is a little different. Do you have similar situation