SSD for single class detection

Question

I\'m training a SSD model to detect faces in images, the maximum amount of faces are 20. I used VGG16 as backbone for the model (with pre-trained weights). after the SSD lay