File size: 9,084 Bytes
f9f785d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
from io import BytesIO

import string
import gradio as gr
import requests
from utils import Endpoint, get_token


def encode_image(image):
    buffered = BytesIO()
    image.save(buffered, format="JPEG")
    buffered.seek(0)

    return buffered


def query_chat_api(
    image, prompt, decoding_method, temperature, len_penalty, repetition_penalty
):

    url = endpoint.url
    url = url + "/api/generate"

    headers = {
        "User-Agent": "BLIP-2 HuggingFace Space",
        "Auth-Token": get_token(),
    }

    data = {
        "prompt": prompt,
        "use_nucleus_sampling": decoding_method == "Nucleus sampling",
        "temperature": temperature,
        "length_penalty": len_penalty,
        "repetition_penalty": repetition_penalty,
    }

    image = encode_image(image)
    files = {"image": image}

    response = requests.post(url, data=data, files=files, headers=headers)

    if response.status_code == 200:
        return response.json()
    else:
        return "Error: " + response.text


def query_caption_api(
    image, decoding_method, temperature, len_penalty, repetition_penalty
):

    url = endpoint.url
    url = url + "/api/caption"

    headers = {
        "User-Agent": "BLIP-2 HuggingFace Space",
        "Auth-Token": get_token(),
    }

    data = {
        "use_nucleus_sampling": decoding_method == "Nucleus sampling",
        "temperature": temperature,
        "length_penalty": len_penalty,
        "repetition_penalty": repetition_penalty,
    }

    image = encode_image(image)
    files = {"image": image}

    response = requests.post(url, data=data, files=files, headers=headers)

    if response.status_code == 200:
        return response.json()
    else:
        return "Error: " + response.text


def postprocess_output(output):
    # if last character is not a punctuation, add a full stop
    if not output[0][-1] in string.punctuation:
        output[0] += "."

    return output


def inference_chat(
    image,
    text_input,
    decoding_method,
    temperature,
    length_penalty,
    repetition_penalty,
    history=[],
):
    text_input = text_input
    history.append(text_input)

    prompt = " ".join(history)

    output = query_chat_api(
        image, prompt, decoding_method, temperature, length_penalty, repetition_penalty
    )
    output = postprocess_output(output)
    history += output

    chat = [
        (history[i], history[i + 1]) for i in range(0, len(history) - 1, 2)
    ]  # convert to tuples of list

    return {chatbot: chat, state: history}


def inference_caption(
    image,
    decoding_method,
    temperature,
    length_penalty,
    repetition_penalty,
):
    output = query_caption_api(
        image, decoding_method, temperature, length_penalty, repetition_penalty
    )

    return output[0]


title = """<h1 align="center">BLIP-2</h1>"""
description = """Gradio demo for BLIP-2, image-to-text generation from Salesforce Research. To use it, simply upload your image, or click one of the examples to load them.
<br> <strong>Disclaimer</strong>: This is a research prototype and is not intended for production use. No data including but not restricted to text and images is collected."""
article = """<strong>Paper</strong>: <a href='https://arxiv.org/abs/2301.12597' target='_blank'>BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models</a>
<br> <strong>Code</strong>: BLIP2 is now integrated into GitHub repo: <a href='https://github.com/salesforce/LAVIS' target='_blank'>LAVIS: a One-stop Library for Language and Vision</a>
<br> <strong>🤗 `transformers` integration</strong>: You can now use `transformers` to use our BLIP-2 models! Check out the <a href='https://huggingface.co/docs/transformers/main/en/model_doc/blip-2' target='_blank'> official docs </a>
<p> <strong>Project Page</strong>: <a href='https://github.com/salesforce/LAVIS/tree/main/projects/blip2' target='_blank'> BLIP2 on LAVIS</a>
<br> <strong>Description</strong>: Captioning results from <strong>BLIP2_OPT_6.7B</strong>. Chat results from <strong>BLIP2_FlanT5xxl</strong>.
"""

endpoint = Endpoint()

examples = [
    ["house.png", "How could someone get out of the house?"],
    ["flower.jpg", "Question: What is this flower and where is it's origin? Answer:"],
    ["pizza.jpg", "What are steps to cook it?"],
    ["sunset.jpg", "Here is a romantic message going along the photo:"],
    ["forbidden_city.webp", "In what dynasties was this place built?"],
]

with gr.Blocks(
    css="""
    .message.svelte-w6rprc.svelte-w6rprc.svelte-w6rprc {font-size: 20px; margin-top: 20px}
    #component-21 > div.wrap.svelte-w6rprc {height: 600px;}
    """
) as iface:
    state = gr.State([])

    gr.Markdown(title)
    gr.Markdown(description)
    gr.Markdown(article)

    with gr.Row():
        with gr.Column(scale=1):
            image_input = gr.Image(type="pil")

            # with gr.Row():
            sampling = gr.Radio(
                choices=["Beam search", "Nucleus sampling"],
                value="Beam search",
                label="Text Decoding Method",
                interactive=True,
            )

            temperature = gr.Slider(
                minimum=0.5,
                maximum=1.0,
                value=1.0,
                step=0.1,
                interactive=True,
                label="Temperature (used with nucleus sampling)",
            )

            len_penalty = gr.Slider(
                minimum=-1.0,
                maximum=2.0,
                value=1.0,
                step=0.2,
                interactive=True,
                label="Length Penalty (set to larger for longer sequence, used with beam search)",
            )

            rep_penalty = gr.Slider(
                minimum=1.0,
                maximum=5.0,
                value=1.5,
                step=0.5,
                interactive=True,
                label="Repeat Penalty (larger value prevents repetition)",
            )

        with gr.Column(scale=1.8):

            with gr.Column():
                caption_output = gr.Textbox(lines=1, label="Caption Output")
                caption_button = gr.Button(
                    value="Caption it!", interactive=True, variant="primary"
                )
                caption_button.click(
                    inference_caption,
                    [
                        image_input,
                        sampling,
                        temperature,
                        len_penalty,
                        rep_penalty,
                    ],
                    [caption_output],
                )

            gr.Markdown("""Trying prompting your input for chat; e.g. example prompt for QA, \"Question: {} Answer:\" Use proper punctuation (e.g., question mark).""")
            with gr.Row():
                with gr.Column(
                    scale=1.5, 
                ):
                    chatbot = gr.Chatbot(
                        label="Chat Output (from FlanT5)",
                    )

                # with gr.Row():
                with gr.Column(scale=1):
                    chat_input = gr.Textbox(lines=1, label="Chat Input")
                    chat_input.submit(
                        inference_chat,
                        [
                            image_input,
                            chat_input,
                            sampling,
                            temperature,
                            len_penalty,
                            rep_penalty,
                            state,
                        ],
                        [chatbot, state],
                    )

                    with gr.Row():
                        clear_button = gr.Button(value="Clear", interactive=True)
                        clear_button.click(
                            lambda: ("", [], []),
                            [],
                            [chat_input, chatbot, state],
                            queue=False,
                        )

                        submit_button = gr.Button(
                            value="Submit", interactive=True, variant="primary"
                        )
                        submit_button.click(
                            inference_chat,
                            [
                                image_input,
                                chat_input,
                                sampling,
                                temperature,
                                len_penalty,
                                rep_penalty,
                                state,
                            ],
                            [chatbot, state],
                        )

            image_input.change(
                lambda: ("", "", []),
                [],
                [chatbot, caption_output, state],
                queue=False,
            )

    examples = gr.Examples(
        examples=examples,
        inputs=[image_input, chat_input],
    )

iface.queue(concurrency_count=1, api_open=False, max_size=10)
iface.launch(enable_queue=True)