Generate ground_truth instance mask using polygons

[txytju]

❓ Questions and Help

When I tried to train mask-rcnn on my own dataset, I found that a well-trained model can't predict good mask on images(in the training set) in which instance polygons are complex(which may have more than 5 polygons). So I'm kind of worried that the complex instance binary masks were not generated correctly. Will anyone tells me that where in the project can I find code that generates binary mask using instance polygons so I can generate ground_truth masks to make sure they are what I exactly want. Thanks in advance!

[fmassa]

Hi,

In order to convert a polygon mask into a binary mask, you can use the .convert method from the Polygon object.
Here is an example of how you can create an instance of the SegmentationMask (which is a list of polygons).

I'm closing this issue as I've replied to your question, but let me know if you have further questions.

[txytju]

Thanks! I use these blocks to plot instance masks and verified that holes in instances are labeled as foreground, which means that binary mask is necessary in my task, as expected in Issue 256