CODEWITHSUNDEEP
objectdeep-learningcomputer-visiondetectionyolo
Abbas Khan
Abbas Khan

Reputation: 66

how to create labels for YOLO-V1 , target bounding box while implementing from scratch?

I'm trying to implement YOLO-V1，but I got some problem when create labels. YOLO-V1 has 2 bounding box, so, when a grid has an object, how to choose which bounding box to put location information? For example: Follow YOLO paper, normalized location of an object in a grid is [0.5, 0.8,.7, 0.9] how to prepare this 1x1x30 tensorenter image description here?? More specifenter image description hereically label for Bounding box-2 ?? This is for labelling not for testing!!! Thank You

Upvotes: 0

Views: 782

Answers (1)

Mostafa Labib
Mostafa Labib

Reputation: 809

The first element in each boundary is 0 or 1 which refers to the existence of an object in this boundary box ( the cell called p in your picture). The loss function changes depending on p according like this: this so if p is 0 it doesn't consider the locations at all.

It's implemented this way to be able to detect overlapping objects in the same box as illustrated in this image here

Upvotes: 1

Related Questions