multi scale evaluation result is different

[beautifulchoi]

Thanks for your great work and I have questions about some result.

I downloaded the pretrained model from your google drive and run "evaluate.py". without multi-scale evaluation implement, the result is quite similar to the paper. however, with it, the result is quite different from the paper. Did I miss something? if not, could you please explain why the result is different?

this is plus v2 model's result(with ms)

this is original v2 model's result(with ms)

and here is what I change

[anamce]

Hi can you please share a gneric evaluation script which can show visualization results as well ?