Dpu on ultra96 v2 pynq

Hi,I am now trying to run mobilenet v2 ssd on ultra96 v2 pynq platform using dpu to achieve acceleration.
These are the output of convolution layers.


ssd_output[0] is the output(score) of each label for each 1917 bounding box.
ssd_output[1] is the encoding output of the coordinate of each 1917 bounding box.
During the coordinate decoding process, a pre-built box called anchor box is needed.
The main problem is that I do not know how to generate those pre-built anchor box with ssd mobilenet v2!
I had seen the dpu-yolo3.ipynb and the pre-built anchor box of yolo3 was given by someone.

Any advice would be helpful!
Tim