search

killthekitten / kaggle-ds-bowl-2018-baseline

Full train/inference/submission pipeline adapted to the competition from https://github.com/matterport/Mask_RCNN
MIT License
103 stars 32 forks source link

About the annotation and mask issues #12

Open wfeng66 opened 2 years ago

wfeng66 commented 2 years ago

Hello Killthekitten,

Thanks for sharing your great work! It would be helpful.

I am working on a project that trains a instance segmentation model by using open image data set. I saw most of people using matterport mask rcnn implementation said need to feed vid format annotation to the model as input. However, the open image data set only comes with images, bitmap masks and some csv files, totally different.

What I am curious about is, it seems that your training is not used with annotation, but with masks. Is it correct？

Thanks a lot!

killthekitten commented 2 years ago

@wfeng66 hey sure. To me annotations and masks are the same thing, they're just synonyms in this context. Having said that, it's been a while since I worked on this, chances are I'm missing something obvious. Have you tried going through readme?

wfeng66 commented 2 years ago

Great! Thank you for your rapid response! I will try your model.