Specifications for Generation of 'image-parse-agnostic-v3.2'

[Atotem]

Hi, following your code, in order to generate the train/test dataset I apparently must prepare the 'image-parse-agnostic-v3.2'. It looks like it's the same file as in the 'openpose_img', but the regions of the neck and torso are dropped to zero values, but, with some peculiar differences. Which is the process to generate this type of data? Why isn't being performed online. Greetings.

Sincerely, @Atotem