retinamask icon indicating copy to clipboard operation
retinamask copied to clipboard

Published 20 hours ago • verified publisher icon chengyangfu

How to train the retina mask on customer data set

Open adeebakausar opened this issue 5 years ago • 3 comments

❓ Questions and Help

adeebakausar avatar Oct 03 '19 09:10 adeebakausar

I am also interested to see if there is an example on how to do this. I have a segmentation dataset with masks for a single class that I'd like to use as training data for this model. Please advise, thanks!

monocongo avatar Jan 18 '20 21:01 monocongo

Hi @monocongo , I follow the COCO setting. If you can format your dataset in COCO format, you should be able to train it. But I haven't updated the code fo r a while. I am worried the new PyTorch code may break the code.

chengyangfu avatar Jan 20 '20 23:01 chengyangfu

@chengyangfu How about the number of classes please? Inside the defaults.py folder, should '_C.MODEL.ROI_BOX_HEAD.NUM_CLASSES =81' be changed to reflect the exact number of classes for the custom dataset or should it be changed to 'Number of classes + 1'? where 1 = background class? In general, how do I handle number of classes when training on my own dataset?

Eldad27 avatar Jan 15 '21 20:01 Eldad27