File size: 8,628 Bytes
1e1438d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a5b5393
 
 
 
 
1e1438d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8ea53b7
1e1438d
 
 
 
 
 
 
 
 
 
 
 
 
e7708f1
1e1438d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8ea53b7
1e1438d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8ea53b7
1e1438d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
import gradio as gr
from PIL import Image
import shutil
import pickle
import random
import json
import os

if not os.path.exists("./data/pacs/"):
    shutil.unpack_archive("./data/pacs.zip", './data/', 'zip')

METHODS = {
    "Textual Inversion (LDM)": "textualinversion_ldm",
    "Textual Inversion (Stable Diffusion)": "none_with_emb_without_multires",
    "DreamBooth": "unet_without_emb_without_multires",
    "Custom Diffusion": "kv_with_emb_without_multires",
}

for method in list(METHODS.values()):
    if not os.path.exists(f"./data/imagenet/images/{method}"):
        shutil.unpack_archive(f"./data/imagenet/images/{method}.zip", f"./", 'zip')
    if not os.path.exists(f"./data/imagenet/compositions/images/{method}"):
        shutil.unpack_archive(f"./data/imagenet/compositions/images/{method}.zip", f"./", 'zip')

method="original"
if not os.path.exists(f"./data/imagenet/images/{method}"):
    shutil.unpack_archive(f"./data/imagenet/images/{method}.zip", f"./", 'zip')

print("Ready to go")


CONCEPTS = {
    "Art Painting": "art_painting",
    "Cartoon": "cartoon",
    "Photo": "photo",
    "Sketch": "sketch",
}

DOMAINS = ["art_painting", "cartoon", "photo", "sketch"]

with open("./data/imagenet/imagenet_mapping.pkl", "rb") as h:
    imagenet_mapping = pickle.load(h)

OBJECTS = []
for k,v in imagenet_mapping.items():
    CONCEPTS[f"{k}:{v}"] = k
    OBJECTS.append(f"{k}:{v}")


def get_domains(method, concept):
    gen_cls=random.choice(os.listdir(os.path.join('./data/pacs', method, concept)))
    fname=random.choice(os.listdir(os.path.join('./data/pacs', method, concept, gen_cls)))
    gen_img = Image.open(os.path.join('./data/pacs', method, concept, gen_cls, fname)).resize((128, 128))

    ref_images = []
    for i in range(3):
        cls=random.choice(os.listdir(os.path.join('./data/pacs', 'original', concept)))
        fname=random.choice(os.listdir(os.path.join('./data/pacs', 'original', concept, cls)))
        img = Image.open(os.path.join('./data/pacs', 'original', concept, cls, fname)).resize((128, 128))
        ref_images.append(img)

    return gen_img, f"a photo of {gen_cls} in the style of {concept}", ref_images


def get_objects(method, concept, evaluation):
    if evaluation=="Concept Alignment":
        gen_cls = ""
        if "ldm" in method:
            gen_cls="samples"
        fname=random.choice(os.listdir(os.path.join('./data/imagenet/images', method, concept, gen_cls)))
        gen_img = Image.open(os.path.join('./data/imagenet/images', method, concept, gen_cls, fname)).resize((128, 128))
        
        ref_images = []
        for i in range(3):
            fname=random.choice(os.listdir(os.path.join('./data/imagenet/images', 'original', concept)))
            img = Image.open(os.path.join('./data/imagenet/images', 'original', concept, fname)).resize((128, 128))
            ref_images.append(img)
        
        return gen_img, f"a photo of **{imagenet_mapping[concept]}**", ref_images

    else:
        gen_cls = ""
        if "ldm" in method:
            gen_cls="samples"

        with open(f"./data/imagenet/compositions/prompts/{concept}.json", "r") as h:
            prompts = json.load(h)
        
        fname=random.choice(os.listdir(os.path.join('./data/imagenet/compositions/images', method, concept, gen_cls)))
        gen_img = Image.open(os.path.join('./data/imagenet/compositions/images', method, concept, gen_cls, fname)).resize((128, 128))
        
        idx = int(fname.split("_")[0])
        caption = prompts[idx]["caption"].replace(prompts[idx]["entity"], f"**{prompts[idx]['entity']}**")

        ref_images = []
        for i in range(3):
            fname=random.choice(os.listdir(os.path.join('./data/imagenet/images', 'original', concept)))
            img = Image.open(os.path.join('./data/imagenet/images', 'original', concept, fname)).resize((128, 128))
            ref_images.append(img)

        return gen_img, caption, ref_images


def get_images(method, concept, evaluation):
    method = METHODS[method]
    concept = CONCEPTS[concept]
    if concept in DOMAINS:
        images, captions, ref_images = get_domains(method, concept)
        return images, captions, ref_images
    elif concept in list(imagenet_mapping.keys()):
        images, captions, ref_images = get_objects(method, concept, evaluation)
        return images, captions, ref_images
    else:
        return


css='''
#image_upload{min-height:4px}
#image_upload [data-testid="image"], #image_upload [data-testid="image"] > div{max-height: 5}
'''
image_blocks = gr.Blocks(css=css)
with image_blocks as demo:

# with gr.Blocks() as demo:
    gr.Markdown("<h1 style='text-align: center;'>ConceptBed Benchmark Explorer</h1>")
    gr.Markdown("<h1 style='text-align: center;'><a href='https://conceptbed.github.io'>Project Page</a> | <a href='https://arxiv.org/abs/2306.04695'>Paper</a>  </h1>")
    gr.Markdown("""
    ## How to interpret results:
    1. The shown three images are reference concept images learned by the diffusion model.
    2. The output target concept image is generated by Stable Diffusion using selected methodologies.
    3. The output text indicates the prompt used to generate the image.
    # """)
    gr.Markdown("""
    ## Types of evaluations:
    1. Concept Alignment: available for all concepts
    2. Compositional Reasoning: available for all concepts except -- Art Painting, Cartoon, Sketch, Photo
    # """)
    gr.Markdown("""
    ### For further details on the ConceptBed benchmark, please refer to the paper at: <a href="https://arxiv.org/abs/2306.04695">https://arxiv.org/abs/2306.04695</a>
    # """)


    with gr.Row():
        with gr.Column():
            methods1 = gr.Dropdown(
                list(METHODS.keys()),
                label="Concept Learner",
                info="Select a concept learning strategy."
                )
            concept1 = gr.Dropdown(
                list(CONCEPTS.keys()),
                label="Concept",
                info="Select a concept."
                )
            evaluation1 = gr.Dropdown(
                ["Concept Alignment", "Compositional Reasoning"], 
                label="Evaluation Type", 
                info="Select the evaluation type."
                )
            gallery1 = gr.Gallery(
                label="Reference images",
                show_label=False,
                elem_id="gallery",
            ).style(
                columns=[3], rows=[1], height="200px"
            )
            # image1 = gr.Gallery(
            #     label="Reference images",
            #     show_label=False,
            #     elem_id="gallery",
            # ).style(
            #     columns=[1], rows=[1], height="200px"
            # )


            image1 = gr.Image()#.style(height="200px", width="200px")
            text1 = gr.Textbox(label="Caption used to generate above image")
            btn1 = gr.Button(value="Get Image", full_width=False)

        with gr.Column():
            methods2 = gr.Dropdown(
                list(METHODS.keys()),
                label="Concept Learner",
                info="Select a concept learning strategy."
                )
            concept2 = gr.Dropdown(
                list(CONCEPTS.keys()),
                label="Concept",
                info="Select a concept."
                )
            evaluation2 = gr.Dropdown(
                ["Concept Alignment", "Compositional Reasoning"], 
                label="Evaluation Type", 
                info="Select the evaluation type."
                )
            gallery2 = gr.Gallery(
                label="Reference images",
                show_label=False,
                elem_id="gallery",
            ).style(
                columns=[3], rows=[1], height="200px"
            )
            image2 = gr.Image(elem_id="image_upload")
            text2 = gr.Textbox(label="Caption used to generate above image")
            btn2 = gr.Button(value="Get Image", full_width=False)


    btn1.click(get_images, inputs=[methods1, concept1, evaluation1], outputs=[image1, text1, gallery1])
    btn2.click(get_images, inputs=[methods2, concept2, evaluation2], outputs=[image2, text2, gallery2])

    with gr.Accordion(label="Notes", open=False):
        gr.HTML(
            """<div class="acknowledgments">
                <p><h4>Generated Images:</h4>
As ConceptBed evaluations required training of 1000+ models (one for each concept), it is impossible to host a live demo.
Therefore, we generate 200,000+ images and randomly select a few images for this demo.
            """
        )

if __name__ == "__main__":
    demo.launch()