search

WXinlong / SOLO

SOLO and SOLOv2 for instance segmentation, ECCV 2020 & NeurIPS 2020.
Other
1.7k stars 306 forks source link

Can I use different S1*S2 where S1≠S2 instead of S*S grids for short fat instances? #216

Open kizoooh opened 2 years ago

kizoooh commented 2 years ago

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!