I am currently training an object segmentation model (detectron2 : mask rcnn)

The objective is to detect materials like wood, plastic, glass etc...

wood is one of the categories in my training set.

Is it a good idea to add other labels that are subcategories of wood? Like table, desk etc ... to my training set? and if so, how should I label them?

Note: I tried adding table, I noticed that the model detects the table as an object and then adds some small detections of wood on top of it. so when labeling the images, should I label the table as table and then label all its parts as wood? or should I just label the whole thing as wood and avoid adding subcategories to my dataset?



You must log in to answer this question.