search

farrukh-x2 / DL-Enthusiasts

A Team of DL Enthusiasts for Kaggle 2018 Data Science Bowl Challenge
5 stars 5 forks source link

Your score is too low #3

Open John1231983 opened 6 years ago

John1231983 commented 6 years ago

Hi, I checked your score team is too low althought you used mask rcnn. Does it score use mask rcnn or unet? If it use maskrcnn so I think you have a wrong in encode submission that need to encode ewch mask individually instead of combined together and encode

farrukh-x2 commented 6 years ago

There is a solution for Unet and Mask RCNN both. It depends on which one you ran. If you are able to increase the score by encoding each mask individually please do share the results.

John1231983 commented 6 years ago

Let check it in https://www.kaggle.com/c/data-science-bowl-2018/discussion/49692

So, what is your LB score use mask rcnn?

farrukh-x2 commented 6 years ago

I think it was above 0.4

John1231983 commented 6 years ago

Him have you try to encode RLE for each mask individually? Does it help to improve your score? I am verg happy if it helped you

farrukh-x2 commented 6 years ago

Can you recommend a method for doing that?

John1231983 commented 6 years ago

Sorry for reply late. You can refer the function https://github.com/YanWang2014/Mask_RCNN/blob/master/Spot-Nuclei2/functions.py#L48