1

I have done a lot of reading on SSD paper, but i couldn't figure out when they mention a scale of Smin 0.2 to Smax 0.9, do they mean in relation to the input image size or in relation to the feature map size at that layer. Can someone help clarify this fact?

Machavity
  • 30,841
  • 27
  • 92
  • 100
Pratik Khadloya
  • 12,509
  • 11
  • 81
  • 106

1 Answers1

0

Apparently this is with respect to each layer, quoting (page 6, equation 4):

The scale of the default boxes for each feature map is computed as:

enter image description here

where smin is 0.2 and smax is 0.9, meaning the lowest layer has a scale of 0.2 and the highest layer has a scale of 0.9, and all layers in between are regularly spaced.

rkellerm
  • 5,362
  • 8
  • 58
  • 95
  • Link to SSD paper: https://arxiv.org/pdf/1512.02325.pdf – Dmitriy Work Jun 19 '21 at 02:59
  • @rkellerm is k the layer number, and m the spatial dimension of the feature map? For example, from the paper, at conv4_3, the size of the feature map is 38x38. Using the formula above, is K=1 and M=38? – rocksyne Feb 10 '23 at 10:23