EADST

Image2Text: Automating Document Layout Analysis with Python and LayoutParser

Image2Text: Automating Document Layout Analysis with Python and LayoutParser

import cv2
import layoutparser as lp
import os
import json
from PIL import Image
import numpy as np

def to_serializable(obj):
    if isinstance(obj, (np.float32, np.float64)):
        return float(obj)
    elif isinstance(obj, np.ndarray):
        return obj.tolist()
    else:
        return obj

def process_image(image_path, model):
    # Read and preprocess the image
    image = cv2.imread(image_path)
    image = image[..., ::-1]  # Convert from BGR to RGB

    # Use the model to detect layout
    layout = model.detect(image)

    # Convert layout objects to a serializable format
    layout_data = []
    for obj in layout:
        obj_dict = obj.to_dict()
        # Iterate through the dictionary, converting all numpy data types to serializable types
        obj_dict_serializable = {key: to_serializable(value) for key, value in obj_dict.items()}
        layout_data.append(obj_dict_serializable)

    return layout_data

def save_layout_to_json(layout_data, json_path):
    # Save layout data to a JSON file
    with open(json_path, 'w') as json_file:
        json.dump(layout_data, json_file)

# Load the model
model = lp.PaddleDetectionLayoutModel(
    config_path="lp://PubLayNet/ppyolov2_r50vd_dcn_365e_publaynet/config",
    threshold=0.5,
    label_map={0: "Text", 1: "Title", 2: "List", 3: "Table", 4: "Figure"},
    enforce_cpu=False,
    enable_mkldnn=True
)

def process_folder(folder_path):
    # Iterate through all files and subfolders in the folder
    for root, dirs, files in os.walk(folder_path):
        for file in files:
            if file.lower().endswith('.jpg'):  # Check if it's a JPG file
                file_path = os.path.join(root, file)
                layout_data = process_image(file_path, model)  # Process the image

                # Create JSON file path
                json_path = os.path.splitext(file_path)[0] + '.json'
                save_layout_to_json(layout_data, json_path)  # Save layout data as JSON


# Specify the folder path to process
folder_path = '/your_folder_path/'
process_folder(folder_path)
相关标签
About Me
XD
Goals determine what you are going to be.
Category
标签云
SVR ms-swift Bipartite MD5 WebCrawler Website torchinfo Logo Augmentation Paper Base64 Crawler uwsgi CTC Qwen Image2Text Land tqdm Miniforge Anaconda Jetson CSV Math Input Conda Google CV mmap NLP 公式 Cloudreve Interview Linux Gemma LaTeX PIP scipy Disk Diagram ModelScope 递归学习法 Streamlit Search ResNet-50 TTS Claude 关于博主 财报 域名 Mixtral FastAPI DeepSeek PyTorch Michelin InvalidArgumentError WAN 报税 论文速读 LLAMA Web XML Dataset 第一性原理 Template Hilton 腾讯云 Knowledge Markdown Heatmap Card Django GPT4 音频 CLAP Translation 论文 News 签证 Pillow C++ DeepStream Magnet transformers Attention Tensor Distillation FP8 飞书 GGML Review API网关 个人博客 Plate icon TSV Food 版权 Safetensors TensorFlow CUDA NLTK VGG-16 图形思考法 Python CC Animate FP64 hf LeetCode git Qwen2.5 Agent SQL Pandas NameSilo SAM 阿里云 LLM OCR Shortcut Password SQLite 强化学习 YOLO 多线程 Random 算法题 RGB AI Sklearn COCO UI logger FP16 uWSGI JSON Permission Tracking Baidu Quantize LoRA RL Video Hotel Zip Data Excel Bin CEIR GoogLeNet API PDB GPTQ 继承 Jev Git Breakpoint Vmess 云服务器 Datetime 搞笑 Nginx OpenCV CAM Statistics HaggingFace HuggingFace tar PDF Firewall Windows EXCEL Numpy QWEN Github Harness Bert Hungarian llama.cpp Ubuntu BTC FP32 v0.dev ONNX Ptyhon Transformers Pickle Use VPN BeautifulSoup Docker 长期记忆 Vim OpenAI IndexTTS2 UNIX 证件照 Bitcoin printf PyCharm Freesound Algorithm Proxy Tiktoken Jupyter RAR SPIE 净利润 Rebuttal Quantization Llama Color Qwen2 Clash 多进程 Plotly 图标 FlashAttention GIT Pytorch 顶会 BF16 Domain TensorRT v2ray ChatGPT Paddle diffusers 自我改进 XGBoost git-lfs VSCode
站点统计

本站现有博文340篇,共被浏览1019230次

本站已经建立2719天!

热门文章
文章归档
回到顶部