Can I use different S1*S2 where S1≠S2 instead of S*S grids for short fat instances?

[kizoooh]

Thank you for your wonderful work!

I'm dealing with short and fat instance segmentations these days, and I found the amount of computation particularly large. I guess the problem is that too much grids are assigned to predict the short fat instances. So can I use rectangle grids instead of square grids to deal with these short fat instances?

I'm new in this area, so I look forward to your reply!