File size: 5,581 Bytes
c3bc979
2773523
c3bc979
2773523
c3bc979
 
 
 
 
 
bd5711f
c3bc979
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bd5711f
 
 
 
 
 
 
c3bc979
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bd5711f
c3bc979
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bd5711f
c3bc979
 
 
 
 
 
 
bd5711f
c3bc979
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from __future__ import annotations

import os

import gradio as gr
import torch
from gradio_client import Client

DESCRIPTION = "# Comparing image captioning models"
ORIGINAL_SPACE_INFO = """\
- [(Omit because it wasn't good)GIT-large fine-tuned on COCO](https://huggingface.co/spaces/library-samples/image-captioning-with-git)
- [BLIP-large](https://huggingface.co/spaces/library-samples/image-captioning-with-blip)
- [BLIP-2 OPT 6.7B](https://huggingface.co/spaces/merve/BLIP2-with-transformers)
- [BLIP-2 T5-XXL](https://huggingface.co/spaces/hysts/BLIP2-with-transformers)
- [InstructBLIP](https://huggingface.co/spaces/library-samples/InstructBLIP)
- [Fuyu-8B](https://huggingface.co/spaces/adept/fuyu-8b-demo)
"""

torch.hub.download_url_to_file("http://images.cocodataset.org/val2017/000000039769.jpg", "cats.jpg")
torch.hub.download_url_to_file(
    "https://huggingface.co/datasets/nielsr/textcaps-sample/resolve/main/stop_sign.png", "stop_sign.png"
)
torch.hub.download_url_to_file(
    "https://cdn.openai.com/dall-e-2/demos/text2im/astronaut/horse/photo/0.jpg", "astronaut.jpg"
)


# def generate_caption_git(image_path: str) -> str:
#     try:
#         client = Client("library-samples/image-captioning-with-git")
#         return client.predict(image_path, api_name="/caption")
#     except Exception:
#         gr.Warning("The GIT-large Space is currently unavailable. Please try again later.")
#         return ""


def generate_caption_blip(image_path: str) -> str:
    try:
        client = Client("library-samples/image-captioning-with-blip")
        return client.predict(image_path, "A picture of", api_name="/caption")
    except Exception:
        gr.Warning("The BLIP-large Space is currently unavailable. Please try again later.")
        return ""


def generate_caption_blip2_opt(image_path: str) -> str:
    try:
        client = Client("merve/BLIP2-with-transformers")
        return client.predict(
            image_path,
            "Beam search",
            1,  # temperature
            1,  # length penalty
            1.5,  # repetition penalty
            api_name="/caption",
        )
    except Exception:
        gr.Warning("The BLIP2 OPT6.7B Space is currently unavailable. Please try again later.")
        return ""


def generate_caption_blip2_t5xxl(image_path: str) -> str:
    try:
        client = Client("hysts/BLIP2-with-transformers")
        return client.predict(
            image_path,
            "Beam search",
            1,  # temperature
            1,  # length penalty
            1.5,  # repetition penalty
            50,  # max length
            1,  # min length
            5,  # number of beams
            0.9,  # top p
            api_name="/caption",
        )
    except Exception:
        gr.Warning("The BLIP2 T5-XXL Space is currently unavailable. Please try again later.")
        return ""


def generate_caption_instructblip(image_path: str) -> str:
    try:
        client = Client("library-samples/InstructBLIP")
        return client.predict(
            image_path,
            "Describe the image.",
            "Beam search",
            5,  # beam size
            256,  # max length
            1,  # min length
            0.9,  # top p
            1.5,  # repetition penalty
            1.0,  # length penalty
            1.0,  # temperature
            api_name="/run",
        )
    except Exception:
        gr.Warning("The InstructBLIP Space is currently unavailable. Please try again later.")
        return ""


def generate_caption_fuyu(image_path: str) -> str:
    try:
        client = Client("adept/fuyu-8b-demo")
        return client.predict(image_path, "Generate a coco style caption.", fn_index=3)
    except Exception:
        gr.Warning("The Fuyu-8B Space is currently unavailable. Please try again later.")
        return ""


def generate_captions(image_path: str) -> tuple[str, str, str, str, str, str]:
    return (
        # generate_caption_git(image_path),
        generate_caption_blip(image_path),
        generate_caption_blip2_opt(image_path),
        generate_caption_blip2_t5xxl(image_path),
        generate_caption_instructblip(image_path),
        generate_caption_fuyu(image_path),
    )


with gr.Blocks(css="style.css") as demo:
    gr.Markdown(DESCRIPTION)
    with gr.Row():
        with gr.Column():
            input_image = gr.Image(type="filepath")
            run_button = gr.Button("Caption")
        with gr.Column():
            # out_git = gr.Textbox(label="GIT-large fine-tuned on COCO")
            out_blip = gr.Textbox(label="BLIP-large")
            out_blip2_opt = gr.Textbox(label="BLIP-2 OPT 6.7B")
            out_blip2_t5xxl = gr.Textbox(label="BLIP-2 T5-XXL")
            out_instructblip = gr.Textbox(label="InstructBLIP")
            out_fuyu = gr.Textbox(label="Fuyu-8B")

    outputs = [
        # out_git,
        out_blip,
        out_blip2_opt,
        out_blip2_t5xxl,
        out_instructblip,
        out_fuyu,
    ]
    gr.Examples(
        examples=[
            "cats.jpg",
            "stop_sign.png",
            "astronaut.jpg",
        ],
        inputs=input_image,
        outputs=outputs,
        fn=generate_captions,
        cache_examples=os.getenv("CACHE_EXAMPLES") == "1",
    )

    with gr.Accordion(label="The original Spaces can be found here:", open=False):
        gr.Markdown(ORIGINAL_SPACE_INFO)

    run_button.click(
        fn=generate_captions,
        inputs=input_image,
        outputs=outputs,
        api_name="caption",
    )

if __name__ == "__main__":
    demo.queue(max_size=20).launch()