search

mrlooi / rotated_maskrcnn

Rotated Mask R-CNN: From Bounding Boxes to Rotated Bounding Boxes
MIT License
347 stars 62 forks source link

Only one box detected per image #15

Closed binwangGit closed 4 years ago

binwangGit commented 4 years ago

❓ I trained this project on ICDAR. And train loss converges to ~0.03, but only one box detected on per trainset image

mrlooi commented 4 years ago

That's odd, are you sure the ICDAR annotation is correct?

Without further examples, I can't really provide a good suggestion

Ivansyf34 commented 1 month ago

❓ I trained this project on ICDAR. And train loss converges to ~0.03, but only one box detected on per trainset image

I have the same question，do you solve it？