search

allanzelener / YAD2K

YAD2K: Yet Another Darknet 2 Keras
Other
2.71k stars 877 forks source link

How anchors size change when you using input image size 608x608? #102

Open mmderakhshani opened 6 years ago

mmderakhshani commented 6 years ago

Hello @allanzelener,

I have got a question about the anchors size. When I look at the anchors size in cfg file of Darknet, They are in the range of 0 to 13 (Because the final feature map size is 13x13 with 416x416 input image size). If I would like to change the input image size for training to 608x608. Should I change the anchors size to the range of 0 to 19? I mean, Should I rescale the anchors size by following formula?

anchors = (anchors / 13) * 19

bicepjai commented 6 years ago

basic question regarding anchor boxes. what are these values relative to, i mean its units

  1. pixels in terms of original image in range [0,416]
  2. [0,1] relative to original image height and width (416)
  3. pixels relative to grid cell size in range [0,32]
  4. [0,13] in terms of pixel
  5. [0,1] relative to grid cell height and width (32 if number of grids is 13)

my question is not related to your question sorry about that ...

raytroop commented 6 years ago

it's relative to final conv feature map