cia-labs / vision-tech

Feral's vision tech part of the pilot program for ISM
0 stars 2 forks source link

Feral's previous labeling strategy #5

Closed Tejas-ChandraShekarRaju closed 3 months ago

Tejas-ChandraShekarRaju commented 11 months ago

What kind of labeling was done previously on the model feral used to train?

  1. Bounding box annotation - which is basically drawing a rectangle where the object is present, in our case where the elephant is present.
  2. Semantic segmentation Labeling - drawing exact boundaries on objects that you want to identify. In our case ear, trunk etc. - Tried in labelMg
  3. Instance segmentation labeling -
  4. Keypoint labeling -
  5. Attribute labeling - labeling age, gender etc. I don’t think it’s that useful for us.
johnkaramchand commented 3 months ago

BLTN; The process was that a human (expert) would look at a picture, label segments "in their mind" and note down the segment's classes in an excel sheet.