# Updated lb position 228 ,unselected 15th place 

It is clearly known factor that there many noisy bbx in the train data unlike test data.
Here are things that worked best for me in 10 days
1) Label correction of small bounding boxes by decreasing the x,y with 2 percent and increasing width ,height by factor of 1.5 . This would have enlarged very point size bbx to some visible level without affecting much accuracy of bboxes
2) Removing all bbx with area less than 500 after above correction
3) Pseudo label training with img size 1024  by removing the noisy train images and usign it along with test images which were far more accurate.
4) 512 img size model worked best on Public LB but it was 1024 that worked best on private LB,it is true in atleast my case.
These above were key things i could think of at best in last 10 days.
I regret i dint select the submission of 1024 img size worked based on above observation and it apparently turned to be one scoring highest in Private lb but moderately on public lb.  67.13 unselected
Few other things due short of time i couldnt explore was Faster RCNN with mobile Net v2 , this was light model could have worked  in this competition.
Congrats to all top positions..
May be if i had joined earlier could have been better placed.
Next Competition is OSIC…