app.py

import pandas as pd
import PIL
from PIL import Image
from PIL import ImageDraw
import gradio as gr
import torch
import easyocr
import omegaconf

from vietocr.model.transformerocr import VietOCR
from vietocr.model.vocab import Vocab
from vietocr.translate import translate, process_input

config = omegaconf.OmegaConf.load("vgg-seq2seq.yaml")
config = omegaconf.OmegaConf.to_container(config, resolve=True)

vocab = Vocab(config['vocab'])
model = VietOCR(len(vocab),
        config['backbone'],
        config['cnn'], 
        config['transformer'],
        config['seq_modeling'])
model.load_state_dict(torch.load('train_old.pth', map_location=torch.device('cpu')))

torch.hub.download_url_to_file('https://github.com/JaidedAI/EasyOCR/raw/master/examples/english.png', 'english.png')
torch.hub.download_url_to_file('https://github.com/JaidedAI/EasyOCR/raw/master/examples/thai.jpg', 'thai.jpg')
torch.hub.download_url_to_file('https://github.com/JaidedAI/EasyOCR/raw/master/examples/french.jpg', 'french.jpg')
torch.hub.download_url_to_file('https://github.com/JaidedAI/EasyOCR/raw/master/examples/chinese.jpg', 'chinese.jpg')
torch.hub.download_url_to_file('https://github.com/JaidedAI/EasyOCR/raw/master/examples/japanese.jpg', 'japanese.jpg')
torch.hub.download_url_to_file('https://github.com/JaidedAI/EasyOCR/raw/master/examples/korean.png', 'korean.png')
torch.hub.download_url_to_file('https://i.imgur.com/mwQFd7G.jpeg', 'Hindi.jpeg')

def draw_boxes(image, bounds, color='yellow', width=2):
    draw = ImageDraw.Draw(image)
    for bound in bounds:
        p0, p1, p2, p3 = bound[0]
        draw.line([*p0, *p1, *p2, *p3, *p0], fill=color, width=width)
    return image

def inference(img, lang):
    reader = easyocr.Reader(lang)
    bounds = reader.readtext(img.name)
    new_bounds=[]
    for (bbox, text, prob) in bounds:
        y0 = bbox[0].min()
        y1 = bbox[0].max()
        x0 = bbox[1].min()
        x1 = bbox[1].max()
        
        # crop the region of interest (ROI)
        img = Image.open(img.name)
        img = img[y0:y1, x0:x1]
        img = process_input(img, config['dataset']['image_height'], 
                    config['dataset']['image_min_width'], config['dataset']['image_max_width'])
        out = translate(img, model)[0].tolist()
        out = vocab.decode(out)
        new_bounds.append(bbox, out, prob)
    im = PIL.Image.open(img.name)
    draw_boxes(im, bounds)
    im.save('result.jpg')
    return ['result.jpg', pd.DataFrame(new_bounds).iloc[: , 1:]]

title = 'EasyOCR'
description = 'Gradio demo for EasyOCR. EasyOCR demo supports 80+ languages.To use it, simply upload your image and choose a language from the dropdown menu, or click one of the examples to load them. Read more at the links below.'
article = "<p style='text-align: center'><a href='https://www.jaided.ai/easyocr/'>Ready-to-use OCR with 80+ supported languages and all popular writing scripts including Latin, Chinese, Arabic, Devanagari, Cyrillic and etc.</a> | <a href='https://github.com/JaidedAI/EasyOCR'>Github Repo</a></p>"
examples = [['english.png',['en']],['thai.jpg',['th']],['french.jpg',['fr', 'en']],['chinese.jpg',['ch_sim', 'en']],['japanese.jpg',['ja', 'en']],['korean.png',['ko', 'en']],['Hindi.jpeg',['hi', 'en']]]
css = ".output_image, .input_image {height: 40rem !important; width: 100% !important;}"
choices = [
    "abq",
    "ady",
    "af",
    "ang",
    "ar",
    "as",
    "ava",
    "az",
    "be",
    "bg",
    "bh",
    "bho",
    "bn",
    "bs",
    "ch_sim",
    "ch_tra",
    "che",
    "cs",
    "cy",
    "da",
    "dar",
    "de",
    "en",
    "es",
    "et",
    "fa",
    "fr",
    "ga",
    "gom",
    "hi",
    "hr",
    "hu",
    "id",
    "inh",
    "is",
    "it",
    "ja",
    "kbd",
    "kn",
    "ko",
    "ku",
    "la",
    "lbe",
    "lez",
    "lt",
    "lv",
    "mah",
    "mai",
    "mi",
    "mn",
    "mr",
    "ms",
    "mt",
    "ne",
    "new",
    "nl",
    "no",
    "oc",
    "pi",
    "pl",
    "pt",
    "ro",
    "ru",
    "rs_cyrillic",
    "rs_latin",
    "sck",
    "sk",
    "sl",
    "sq",
    "sv",
    "sw",
    "ta",
    "tab",
    "te",
    "th",
    "tjk",
    "tl",
    "tr",
    "ug",
    "uk",
    "ur",
    "uz",
    "vi"
]
gr.Interface(
    inference,
    [gr.inputs.Image(type='file', label='Input'),gr.inputs.CheckboxGroup(choices, type="value", default=['en'], label='language')],
    [gr.outputs.Image(type='file', label='Output'), gr.outputs.Dataframe(headers=['text', 'confidence'])],
    title=title,
    description=description,
    article=article,
    examples=examples,
    css=css,
    enable_queue=True
    ).launch(debug=True)