Object detection

I have some confusions about how SSD object detection algorithm works.

  1. Where the anchor boxes are applied? (Only to the ground truth boxes or predictions also)
  2. How the loss function is calculated
  3. Why did Jeremy chose number of output as 4x4x(4+c)

Here’s a very detailed and great write up about SSD, hopefully you’ll find your answers there ! :slight_smile:


Thanks a lot:)