app.py

import gradio as gr
from llava.eval.model_vqa_demo import eval_model
import os

def inference(my_prompt, my_image):
    outputs = eval_model(my_image, my_prompt)
    return outputs

with gr.Blocks(css=".custom-title { text-align: center; font-size: 2rem; font-weight: bold; margin-bottom: 20px; }") as demo:
    with gr.Row():
        gr.Markdown("<div class='custom-title'>VLC Demo</div>")
    gr.Markdown("This is the official demo of the latest version of 'Vision-Language-Camera: Aligning User Intent and Camera Expertise through Vision-Language Models'")
    with gr.Row():
        with gr.Column(scale=1):
            prompt = gr.Textbox(label="Your Prompt", placeholder="Type your description or question...", lines=3)
            image = gr.Image(label="Input Image", type='filepath')
            run_btn = gr.Button("Run Inference", variant="primary")
            cur_dir = os.path.dirname(os.path.abspath(__file__))
            gr.Examples(
                examples=[
                    [
                        f"{cur_dir}/examples/img1.jpg",
                        "How to tune the camera parameters to make this image clearer?",
                    ],
                ],
                    inputs=[image, prompt],
                )
        with gr.Column(scale=2):
            output_text = gr.Textbox(label="VLC Response", lines=3)

    run_btn.click(fn=inference, inputs=[prompt, image], outputs=[output_text])

demo.launch()