|
|
|
import gradio as gr |
|
import numpy as np |
|
import requests |
|
import base64 |
|
import os |
|
|
|
API_ENDPOINT = os.getenv('API_ENDPOINT') |
|
API_KEY = os.getenv('API_KEY') |
|
|
|
title = "<h1><center>Markup-to-Image Diffusion Models with Scheduled Sampling</center></h1>" |
|
authors = "<center>Yuntian Deng, Noriyuki Kojima, Alexander M. Rush</center>" |
|
info = '<center><a href="https://arxiv.org/pdf/2210.05147.pdf">Paper</a> <a href="https://github.com/da03/markup2im">Code</a></center>' |
|
|
|
with gr.Blocks() as demo: |
|
gr.Markdown(title) |
|
gr.Markdown(authors) |
|
gr.Markdown(info) |
|
with gr.Row(): |
|
with gr.Column(scale=2): |
|
textbox = gr.Textbox(label=r'Type LaTeX formula below and click "Generate"', lines=1, max_lines=1, placeholder='Type LaTeX formula here and click "Generate"', value=r'\sum_{t=1}^T\E_{y_t \sim {\tilde P(y_t| y_0)}} \left\| \frac{y_t - \sqrt{\bar{\alpha}_t}y_0}{\sqrt{1-\bar{\alpha}_t}} - \epsilon_\theta(y_t, t)\right\|^2.') |
|
submit_btn = gr.Button("Generate", elem_id="btn") |
|
with gr.Column(scale=3): |
|
slider = gr.Slider(0, 1000, value=0, label='step (out of 1000)') |
|
image = gr.Image(label="Rendered Image", show_label=False, elem_id="image") |
|
inputs = [textbox] |
|
outputs = [slider, image, submit_btn] |
|
def infer(formula): |
|
data = {'formula': formula, 'api_key': API_KEY} |
|
try: |
|
with requests.post(url=API_ENDPOINT, data=data, timeout=600, stream=True) as r: |
|
i = 0 |
|
for line in r.iter_lines(): |
|
response = line.decode('ascii').strip() |
|
r = base64.decodebytes(response.encode('ascii')) |
|
q = np.frombuffer(r, dtype=np.float32).reshape((64, 320, 3)) |
|
i += 1 |
|
yield i, q, submit_btn.update(visible=False) |
|
yield i, q, submit_btn.update(visible=True) |
|
except Exception as e: |
|
yield 1000, 255*np.ones((64, 320, 3)), submit_btn.update(visible=True) |
|
submit_btn.click(fn=infer, inputs=inputs, outputs=outputs) |
|
demo.queue(concurrency_count=20, max_size=200).launch(enable_queue=True) |
|
|
