Search code examples
conv-neural-networkobject-detectionyolo

What does scale and step values in .cfg file of YOLO means?

I am trying to understand the .cfg file of YOLOv2. I didn't understand

steps=-1,100,80000,100000

scales=.1,10,.1,.1

Can someone explain me this.

Solution

  • steps is a checkpoints (number of iterations) at which scales will be applied.

    scales is a coefficients at which learning_rate will be multiplied at this checkpoints. Determines how the learning_rate will be changed during increasing number of iterations during training.

    Both of them are related to each other and they have the same amount.

    steps=200,400,600,20000,30000

scales=2.5,2,2,.1,.1

    At step 200 the scales is 2.5, step 400 the scale is 2, etc.

    https://github.com/AlexeyAB/darknet/issues/279