app.py

import onnxruntime as onr
import numpy as np
import gradio as gr
import glob, io
from PIL import Image
from cairosvg import svg2png
#----- CONFIG ------
img_height = 300
img_width = 500
max_length = 4
characters = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'A', 'B', 'C', 'D', 'E', 
              'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 
              'U', 'V', 'W', 'X', 'Y', 'Z', 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 
              'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 
              'y', 'z'
              ]
model_path = 'model.onnx'
#===== some code init =====
Model = onr.InferenceSession(model_path)

#===== some funcs =====
def get_result(pred):
        accuracy = 1
        last = None
        ans = []
        for item in pred[0]:
            char_ind = item.argmax()
            if char_ind != last and char_ind != 0 and char_ind != len(characters) + 1:
                ans.append(characters[char_ind - 1])
                accuracy *= item[char_ind]
            last = char_ind
        answ = "".join(ans)[:max_length]
        return answ, accuracy
def predict(svgdata):
    img = Image.open(io.BytesIO(svg2png(svgdata)))
    img = img.convert('L')
    img = img.resize((img_width, img_height))
    img = np.array(img)
    img = np.expand_dims(img, axis=1)
    img = np.expand_dims(img, axis=-1)
    img = img.transpose([1,2,0,3])
    img = img.astype(np.float32) / 255.
    result_tensor = Model.run(None, {'image': img, 'label': np.random.default_rng().random((28, 28), dtype=np.float32)})[0]
    return (get_result(result_tensor)[0])

title = "Made with love❤️❤️❤️❤️\nby Vermei (for userscript)"
description = "thanks to HF for allowing captcha solve userscirpts (code 2)"

iface = gr.Interface(fn=predict, 
                     inputs=gr.inputs.Textbox(), 
                     outputs=gr.outputs.Textbox(),
                     title=title,
                     #examples=glob.glob('examples/*.png'),
                     description=description)
iface.launch()