# coding: utf-8 # [Pix2Text](https://github.com/breezedeus/pix2text): an Open-Source Alternative to Mathpix. # Copyright (C) 2022-2024, [Breezedeus](https://www.breezedeus.com). import os import json import functools import random import shutil import string import tempfile import time import zipfile from pathlib import Path import yaml import gradio as gr import numpy as np from huggingface_hub import hf_hub_download # from cnstd.utils import pil_to_numpy, imsave from pix2text import Pix2Text from pix2text.utils import set_logger, merge_line_texts logger = set_logger() LANGUAGES = yaml.safe_load(open('languages.yaml', 'r', encoding='utf-8'))['languages'] OUTPUT_RESULT_DIR = Path('./output-results') OUTPUT_RESULT_DIR.mkdir(exist_ok=True) def prepare_mfd_model(): target_fp = './yolov7-model/mfd-yolov7-epoch224-20230613.pt' if os.path.exists(target_fp): return target_fp HF_TOKEN = os.environ.get('HF_TOKEN') local_path = hf_hub_download( repo_id='breezedeus/paid-models', subfolder='cnstd/1.2', filename='yolov7-model-20230613.zip', repo_type="model", cache_dir='./', token=HF_TOKEN, ) with zipfile.ZipFile(local_path) as zf: zf.extractall('./') return target_fp def get_p2t_model(lan_list: list, mfd_model_name: str, mfr_model_name: str): analyzer_config = {} if 'yolov7_tiny' not in mfd_model_name: mfd_fp = prepare_mfd_model() analyzer_config = dict( # 声明 LayoutAnalyzer 的初始化参数 model_name='mfd', model_type='yolov7', # 表示使用的是 YoloV7 模型,而不是 YoloV7_Tiny 模型 model_fp=mfd_fp, # 注:修改成你的模型文件所存储的路径 ) formula_config = {} if 'mfr-pro' in mfr_model_name: formula_config = dict( # 声明 LayoutAnalyzer 的初始化参数 model_name='mfr-pro', model_backend='onnx', ) p2t = Pix2Text( languages=lan_list, analyzer_config=analyzer_config, formula_config=formula_config, ) return p2t def latex_render(latex_str): return f"$$\n{latex_str}\n$$" # return latex_str def recognize( lang_list, mfd_model_name, mfr_model_name, rec_type, resized_shape, image_file ): lang_list = [LANGUAGES[l] for l in lang_list] p2t = get_p2t_model(lang_list, mfd_model_name, mfr_model_name) if rec_type == 'mixed': suffix = list(string.ascii_letters) random.shuffle(suffix) suffix = ''.join(suffix[:6]) out_det_fp = f'out-det-{time.time()}-{suffix}.jpg' # 如果 OUTPUT_RESULT_DIR 文件数量超过 1000,按时间删除最早的 1000 个文件 if len(os.listdir(OUTPUT_RESULT_DIR)) > 1000: for fp in sorted(os.listdir(OUTPUT_RESULT_DIR))[:1000]: os.remove(OUTPUT_RESULT_DIR / fp) outs = p2t.recognize( image_file, resized_shape=resized_shape, save_analysis_res=OUTPUT_RESULT_DIR / out_det_fp, ) # To get just the text contents, use: only_text = merge_line_texts(outs, auto_line_break=True) # return only_text, latex_render(only_text) return only_text, str(OUTPUT_RESULT_DIR / out_det_fp) elif rec_type == 'formula': only_text = p2t.recognize_formula(image_file) return latex_render(only_text), './docs/no-det-res.jpg' elif rec_type == 'text': only_text = p2t.recognize_text(image_file) return only_text, './docs/no-det-res.jpg' def example_func(lang_list, rec_type, image_file): return recognize( lang_list, mfd_model_name='yolov7 (paid)', mfr_model_name='mfr-pro (paid)', rec_type=rec_type, resized_shape=768, image_file=image_file, ) def main(): langs = list(LANGUAGES.keys()) langs.sort(key=lambda x: x.lower()) title = ': a Free Alternative to Mathpix' examples = [ [ ['English'], 'mixed', 'docs/examples/mixed-en.jpg', ], [ ['English', 'Chinese Simplified'], 'mixed', 'docs/examples/mixed-ch_sim.jpg', ], [ ['English', 'Chinese Traditional'], 'mixed', 'docs/examples/mixed-ch_tra.jpg', ], [ ['English', 'Vietnamese'], 'mixed', 'docs/examples/mixed-vietnamese.jpg', ], [ ['English'], 'formula', 'docs/examples/formula1.png' ], [ ['English'], 'formula', 'docs/examples/formula2.jpg' ], [ ['English'], 'formula', 'docs/examples/hw-formula.png' ], [ ['English', 'Chinese Simplified'], 'text', 'docs/examples/pure-text.jpg', ], ] table_desc = """