app.py

import os
import datetime
import json
import base64
from PIL import Image
import gradio as gr
import hashlib
import requests
from utils import build_logger
import io

LOGDIR = "log"
logger = build_logger("otter", LOGDIR)

# no_change_btn = gr.Button.update()
# enable_btn = gr.Button.update(interactive=True)
# disable_btn = gr.Button.update(interactive=False)


def decode_image(encoded_image: str) -> Image:
    decoded_bytes = base64.b64decode(encoded_image.encode("utf-8"))
    buffer = io.BytesIO(decoded_bytes)
    image = Image.open(buffer)
    return image


def encode_image(image: Image.Image, format: str = "PNG") -> str:
    with io.BytesIO() as buffer:
        image.save(buffer, format=format)
        encoded_image = base64.b64encode(buffer.getvalue()).decode("utf-8")
    return encoded_image


def get_conv_log_filename():
    t = datetime.datetime.now()
    name = os.path.join(LOGDIR, f"{t.year}-{t.month:02d}-{t.day:02d}-conv.json")
    return name


def get_conv_image_dir():
    name = os.path.join(LOGDIR, "images")
    os.makedirs(name, exist_ok=True)
    return name


def get_image_name(image, image_dir=None):
    buffer = io.BytesIO()
    image.save(buffer, format="PNG")
    image_bytes = buffer.getvalue()
    md5 = hashlib.md5(image_bytes).hexdigest()

    if image_dir is not None:
        image_name = os.path.join(image_dir, md5 + ".png")
    else:
        image_name = md5 + ".png"

    return image_name


def resize_image(image, max_size):
    width, height = image.size
    aspect_ratio = float(width) / float(height)

    if width > height:
        new_width = max_size
        new_height = int(new_width / aspect_ratio)
    else:
        new_height = max_size
        new_width = int(new_height * aspect_ratio)

    resized_image = image.resize((new_width, new_height))
    return resized_image


def http_bot(image_input, text_input, request: gr.Request):
    logger.info(f"http_bot. ip: {request.client.host}")
    print(f"Prompt request: {text_input}")

    base64_image_str = encode_image(image_input)

    payload = {
        "content": [
            {
                "prompt": text_input,
                "image": base64_image_str,
            }
        ],
        "token": "sk-OtterHD",
    }

    print(
        "request: ",
        {
            "prompt": text_input,
            "image": base64_image_str[:10],
        },
    )

    url = "https://rouge-surrey-katrina-signatures.trycloudflare.com/app/otter"
    headers = {"Content-Type": "application/json"}

    response = requests.post(url, headers=headers, data=json.dumps(payload))
    results = response.json()
    print("response: ", {"result": results["result"]})
    return results["result"]


title = """
# OTTER-HD: A High-Resolution Multi-modality Model
[[Otter Codebase]](https://github.com/Luodian/Otter) [[Paper]](https://arxiv.org/abs/2311.04219) [[Checkpoints & Benchmarks]](https://huggingface.co/Otter-AI) 

**OtterHD** is a multimodal fine-tuned from [Fuyu-8B](https://huggingface.co/adept/fuyu-8b) to facilitate a more fine-grained interpretation of high-resolution visual input *without a explicit vision encoder module*. All image patches are linear transformed and processed together with text tokens. This is a very innovative and elegant exploration. We are fascinated and paved in this way, we opensourced the finetune script for Fuyu-8B and improve training throughput by 4-5 times faster with [Flash-Attention-2](https://github.com/Dao-AILab/flash-attention).

**Tips**: 
- Since high-res images are large that may cause the longer transmit time from HF Space to our backend server. Please be kinda patient for the response.
- The model is currently mainly focus on high-res image resolution and need to be futher improved on (1) hallucination reduction (2) text formatting control and some more you can spot and suggest to us.
- We are working on to finetune the model on LLaVA-1.5/LRV/LLaVAR data mixture and balance the detailed recognition and hallucination reduction. Stay tuned!
- Please do not upload any NSFW images and ask relevant questions. We will ban the IP address if we found any inappropriate usage.
"""

css = """
  #mkd {
    height: 1000px; 
    overflow: auto; 
    border: 1px solid #ccc; 
  }
"""

if __name__ == "__main__":
    with gr.Blocks(css=css) as demo:
        gr.Markdown(title)
        dialog_state = gr.State()
        input_state = gr.State()
        with gr.Tab("Ask a Question"):
            with gr.Row(equal_height=True):
                with gr.Column(scale=2):
                    image_input = gr.Image(label="Upload a High-Res Image", type="pil")
                with gr.Column(scale=1):
                    vqa_output = gr.Textbox(label="Output")
            text_input = gr.Textbox(label="Ask a Question")

            vqa_btn = gr.Button("Send It")

            gr.Examples(
                [
                    [
                        "./assets/IMG_00095.png",
                        "How many camels are inside this image?",
                    ],
                    [
                        "./assets/IMG_00057.png",
                        "What's this image about?",
                    ],
                    [
                        "./assets/IMG_00040.png",
                        "What are the scene texts in this image?",
                    ],
                    [
                        "./assets/./IMG_00012.png",
                        "How many apples are there? Count them row by row.",
                    ],
                    [
                        "./assets/IMG_00080.png",
                        "What is this and where is it from?",
                    ],
                    [
                        "./assets/IMG_00041.png",
                        "What are the scene texts in this image?",
                    ],
                ],
                inputs=[image_input, text_input],
                outputs=[vqa_output],
                fn=http_bot,
                label="Click on any Examples below👇",
            )
        vqa_btn.click(fn=http_bot, inputs=[image_input, text_input], outputs=vqa_output)

    demo.launch()