Results un-reproducible after your revision on Nov 6, 2019

yanxp / MetaR-CNN

Meta R-CNN : Towards General Solver for Instance-level Low-shot Learning

https://yanxp.github.io/metarcnn.html

177 stars 23 forks source link

Results un-reproducible after your revision on Nov 6, 2019 #19

Open Ze-Yang opened 4 years ago

Ze-Yang commented 4 years ago

As I report in the previous issue, I notice that you have fixed it. However, this will lead to significant decrease of the performance due to the reduce of annotations, e.g., VOC setup1 5shot, dropping from 43.5% to 36.5%. Could you explain it? Otherwise, I regard that the promising results in your paper maybe just owe to the bug you fixed.

yanxp commented 4 years ago

hello, the following code is our original implementation. We thought they are equivalent. And we re-run the setup1 5shot , we can get the result as 45.1%. Screenshot from 2019-11-19 10-39-48

Ze-Yang commented 4 years ago

Could you reproduce the result with your latest code? I rerun your code for many times, but cannot get the result you indicate. Thanks.

deepseawhale commented 2 years ago

hello, the following code is our original implementation. We thought they are equivalent. And we re-run the setup1 5shot , we can get the result as 45.1%.

Thanks for pointing this, may I know the corresponding commit revision please?