Logical question from paper

[vaishnavm217]

Hi, I have been working on the implementation of faceboxes in pytorch. The most confusing part is the densification part for me. So output of first layer is 21 anchors. So for 32 scale they have 16 anchors, but how is each anchor steps defined?

like for 32 its written 32 steps between each anchor, but that makes density low, so they densify it 4 (or 16?) times but at what steps?

[vaishnavm217]

I went through your code, and all 16 anchor boxes are equally spaced? So that means they are repeated? https://github.com/TropComplique/FaceBoxes-tensorflow/blob/1da7d8dd3dc816eeaf36082c9679a4be2fb70b11/src/anchor_generator.py#L165-L186 Here you are repeating the offsets for the whole n² anchors https://github.com/TropComplique/FaceBoxes-tensorflow/blob/1da7d8dd3dc816eeaf36082c9679a4be2fb70b11/src/anchor_generator.py#L133-L136

And here is the repeated anchor centers. So same anchors are just repeated, n² times?

[TropComplique]

Hi. An anchor just repeated n^2 times near its original center. I mean, it slightly jittered about its original position. That's how I understand it. See the Figure 3 in the original paper.

Please run the following notebook: https://github.com/TropComplique/FaceBoxes-tensorflow/blob/master/visualize_densified_anchor_boxes.ipynb It visualizes the generated anchors.