AlexeyAB / Yolo_mark

GUI for marking bounded boxes of objects in images for training neural network Yolo v3 and v2

Home Page:https://github.com/AlexeyAB/darknet

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Do I need to mark all objects in an image

nathnim opened this issue · comments

A have images database with a lot of objects in the same class in an image, for example a lot of motorbikes. Do I need to mark all objects? If I mark only one object and use such kind of images to train. Does it affect to the model quality? Thanks!

Do I need to mark all objects?

Yes, you must.

If I mark only one object and use such kind of images to train. Does it affect to the model quality?

The network will learn not to detect non-marked bikes.
DNN will distinguish marked and non-marked bikes(will look for features, what distinguishes a marked bike), and will only detect bikes that are similar to marked ones, and will not detect similar to unmarked ones.

Thank you very much! That why even I have 20k of images for an class but the quality of the model which I built is very low. I will try to improve!