File size: 2,443 Bytes
1f1b97c
 
 
 
 
bb63fba
 
 
 
 
 
 
 
 
1f1b97c
0dd7b09
 
 
 
6352ff4
 
1f1b97c
0dd7b09
 
 
6352ff4
 
1f1b97c
 
 
0dd7b09
1f1b97c
 
0dd7b09
 
1f1b97c
 
0dd7b09
 
6352ff4
1f1b97c
0dd7b09
 
6352ff4
1f1b97c
0dd7b09
1f1b97c
0dd7b09
 
1f1b97c
d79806b
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
import gradio as gr
import random, os, shutil
from PIL import Image
import pandas as pd
import tempfile
import zipfile

with zipfile.ZipFile("images/fair_diffusion.zip","r") as zip_ref:
    zip_ref.extractall("images/")


with zipfile.ZipFile("images/stable_diffusion.zip","r") as zip_ref:
    zip_ref.extractall("images/")


   
def open_stable_ims(profession):
    if len(profession) != 0:
        dirname = 'images/stable_diffusion/'+ profession+'/'
        images = random.shuffle([Image.open(os.path.join(dirname+im)).convert("RGB") for im in os.listdir(dirname)])
        return images[:16]

def open_fair_ims(profession):
    if len(profession) != 0:
        dirname = 'images/fair_diffusion/' + profession+'/'
        images = random.shuffle([Image.open(os.path.join(dirname+im)).convert("RGB") for im in os.listdir(dirname)])
        return images[:16]



professions = sorted(os.listdir('images/fair_diffusion'))

with gr.Blocks() as demo:
    gr.Markdown("# Fair Diffusion Explorer")
    gr.Markdown("#### Choose from the occupations below to compare how Stable Diffusion (left) and Fair Diffusion (right) represent different professions.")
    with gr.Row():
        with gr.Column():
            gr.Markdown('## Stable Diffusion Generations')
            choice1 = gr.Dropdown(professions, label = "Choose a profession", multiselect= False, interactive=True)
            images1 = gr.Gallery(label="Images").style(grid=[4], height="auto")
        with gr.Column():
            gr.Markdown('## Fair Diffusion Generations')
            choice2 = gr.Dropdown(professions, label = "Choose a profession", multiselect = False, interactive=True)
            images2 = gr.Gallery(label="Images").style(grid=[4], height="auto")

    gr.Markdown("We present a novel strategy, called **Fair Diffusion**, to attenuate biases after the deployment of generative text-to-image models. Specifically, we demonstrate shifting a bias, based on human instructions, in any direction yielding arbitrarily new proportions for, e.g., identity groups. As our empirical evaluation demonstrates, this introduced control enables instructing generative image models on fairness, with no data filtering and additional training required. For the full paper by Friedrich  et al., see [here](https://arxiv.org/pdf/2302.10893.pdf).")

    choice1.change(open_stable_ims, choice1, [images1])
    choice2.change(open_fair_ims, choice2, [images2])

    demo.launch()