search

matterport / Mask_RCNN

Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
Other
24.55k stars 11.69k forks source link

Annotation of complex shapes / Polygonal holes #1837

Open noamnav opened 4 years ago

noamnav commented 4 years ago

In the below example, I am trying to annotate the 'grass' class. What would be the best technique in this instance? Naturally, a simple polygon would have to include the sheep objects which is undesired.

Thanks

4782628554_668bc31826_z

suchiz commented 4 years ago

Might not be the best solution but what I would do in your case is contouring the sheep, contouring the grass. Then you'll obtain two different masks where you can use binary operator. I think in your case is:

final_grass = grass_with_sheeps and not sheep

Well, it's been 15days already, you probably already found something better.

noamnav commented 4 years ago

That's definitely an option and has been suggested here.

I am however trying a different approach as suggested by @waleedka. results so far are within reason but far from ideal so trying to expand my training set.

There is perhaps a greater question to be asked, which is if the implemented method is at all appropriate for detecting 'background' objects. This could be anything like the grass example above, sky images, roads (separated from cars) or even walls / tiles in interior images (distinguished from 'noise' elements).