app.py

import os
import cv2
import torch
import numpy as np
from PIL import Image
import torchvision.transforms as T
from torch.autograd import Variable
from huggingface_hub import hf_hub_download
import streamlit as st

TOKEN = os.getenv('hf_read_token')
repo_id = "Nischay103/captcha_recognition"
model_files = {
    "v1": "captcha_model_v1.pt", "v2": "captcha_model_v2.pt",
    "v3": "captcha_model_v3.pt", "v4": "captcha_model_v4.pt",
    "v5": "captcha_model_v5.pt", "v6": "captcha_model_v6.pt",
    "v7": "captcha_model_v7.pt", "v8": "captcha_model_v8.pt",
}

example_captchas = {
    "v1": "v1/v1_MYCbSs.jpg", "v2": "v2/v2_032891.png", "v3": "v3/v3_5Bg5m.png",
    "v4": "v4/v4_e78d97.jpg", "v5": "v5/v5_DPMTZ.png", "v6": "v6/v6_WD7A.png",
    "v7": "v7/v7_897375.png", "v8": "v8/v8_qxv6x.png",
}

_decode_cls_dims = {
    "63": 63, "37_uppercase": 37,
    "11": 11, "37_lowercase": 37
}

len_dim_pair = {
    "v1": (6, "63"), "v2": (6, "11"), "v3": (5, "63"),
    "v4": (6, "37_lowercase"), "v5": (5, "37_uppercase"),
    "v6": (4, "63"), "v7": (6, "11"), "v8": (5, "37_uppercase")
}

char_sets = {
    "63": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789$",
    "37_uppercase": "ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789$",
    "37_lowercase": "abcdefghijklmnopqrstuvwxyz0123456789$",
    "11": "0123456789$"
}

models = {}
for key, model_file in model_files.items():
    model_path = hf_hub_download(repo_id=repo_id, filename=model_file, token=TOKEN)
    models[key] = torch.jit.load(model_path)

def transform_image(image_path):
    transform = T.Compose([T.ToTensor()])
    device = 'cuda' if torch.cuda.is_available() else 'cpu'
    image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
    image = Image.fromarray(image)
    image = image.resize((200, 50))
    image = transform(image)
    image = Variable(image).to(device)
    image = image.unsqueeze(1)
    return image

def get_label(model_prediction, model_version):
    max_captcha_len, cls_dim_encoded = len_dim_pair[model_version]
    _cls = char_sets[cls_dim_encoded]
    cls_dim = _decode_cls_dims[cls_dim_encoded]
    lab = ""
    for idx in range(max_captcha_len):
        start = cls_dim * idx
        end = cls_dim * (idx + 1)
        get_char = _cls[torch.argmax(model_prediction[0, start:end])]
        lab += get_char
    return lab

st.title("char-seq recognition from scene-image (captcha)")
st.write("recognize captchas using different models")

uploaded_file = st.file_uploader("choose a captcha image...", type=["jpg", "png"])
model_version = st.selectbox("model variant", list(model_files.keys()), index=0)

if uploaded_file is not None:
    with open("temp_captcha_image.png", "wb") as f:
        f.write(uploaded_file.getbuffer())
    
    input_image_path = "temp_captcha_image.png"
    st.image(input_image_path, caption='uploaded captcha image', use_column_width=True)
    
    if st.button('recognize'):
        input = transform_image(input_image_path)
        model = models[model_version]
        with torch.no_grad():
            model_prediction = model(input)
            output = get_label(model_prediction, model_version)
        st.write(f"Recognized Character Sequence: {output}")

st.write("## examples")
cols = st.columns(4)
for idx,(model_variant,captcha_path) in enumerate(example_captchas.items()):
    col = cols[idx % 4]
    col.image(captcha_path,caption=f'{captcha_path.split("/")[-1]}', width=100)
    col.write(f"Model Version: {model_variant}")