app.py

import os,torch
import gradio as gr
from PIL import Image
from torchvision import transforms as T
from huggingface_hub import hf_hub_download
from tokenizer import Tokenizer

TOKEN = os.getenv('hf_read_token')
repo_id    = "Nischay103/captcha_recognition"
model_file = 'captcha_model.pt'
model_path = hf_hub_download(repo_id=repo_id, filename=model_file, token=TOKEN)
model = torch.jit.load(model_path)

charset = r"0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ"
_tokenizer = Tokenizer(charset)

def get_transform(img_size=(32,128)):
    transforms = []
    transforms.extend([
        T.Resize(img_size, T.InterpolationMode.BICUBIC),
        T.ToTensor(),
        T.Normalize(0.5, 0.5)
    ])
    return T.Compose(transforms)

transform = get_transform()
image = gr.Image(type="filepath", label="Captcha Image")
text = gr.Textbox(label="Recognized Character Sequence")


def recognize_captcha(image_path):
    input = Image.open(image_path)
    input = transform(input.convert('RGB')).unsqueeze(0)
    model = model
    with torch.no_grad():
         probs   = model(input)
         preds,_ =  _tokenizer.decode(probs,beam_width=2)
    return preds[0]

iface = gr.Interface(
    fn=recognize_captcha,
    inputs=image,
    outputs=text,
    title="Character Sequence Recognition from Scene-Image (Captcha)",
    description="Recognize captchas using different models",
    examples=['examples/251615.png']
)

iface.launch()