Multi-tag task

[jackskr666]

tags:

name: Bangs
tag_irrelevant_conditions_dim: 2
attributes: 
  -
    name: 'with'
    filename: datasets/Bangs_with.txt
  -
    name: 'without'
    filename: datasets/Bangs_without.txt

• name: Eyeglasses
  tag_irrelevant_conditions_dim: 2
  attributes:

  name: 'with'
  filename: datasets/Eyeglasses_with.txt
  

  • name: 'without'
    filename: datasets/Eyeglasses_without.txt

if I want to make experiment on Multi-tag task, how to replace Multi-tag task on celeba-hq.yaml setting?

[Xinyang Li]

Hi, here are the steps:

1. please refer to (https://github.com/imlixinyang/HiSD/blob/main/preprocessors/celeba-hq.py) and modify it for your desired tags and attributes. You will get some datasets/Tag_attribute.txt files if succeed.
2. modify the config file to use the your own datasets/Tag_attribute.txt files.
3. train.

[jackskr666]

if I changing the style of hair while changing the style of glasses, so my custom.yaml modify follow the settings?

tags:

name: Eyeglasses_black
tag_irrelevant_conditions_dim: 2
attributes: 
  -
    name: 'with'
    filename: datasets/Eyeglasses_with.txt
  -
    name: 'without'
    filename: datasets/Eyeglasses_without.txt
  -
    name: 'black'
    filename: datasets/HairColor_black.txt

[Xinyang Li]

This config file is only for training. If you want to edit multiple tags in inference, you should edit code in here.

[jackskr666]

Thank you very much, I find facial age transformation in supplementary. How to generate age 7-9’/‘10-14’/‘15-19’/‘20-29’/‘30-39’/‘40-49’/‘50-69 in test.py?

[Xinyang Li]

Hi @jackskr666, sorry for the late reply.
For age transformation, we use the images from FFHQ dataset and the labels provided by FFHQ-aging. You may process them into the label structure of HiSD to reproduce this experiment.