PaddleX/zhch/test_ppstructure_v3.py

from paddlex import create_pipeline
import paddle
import time
from pathlib import Path
import os
from typing import List

from cuda_utils import detect_available_gpus, monitor_gpu_memory

from dotenv import load_dotenv
load_dotenv(override=True)


if __name__ == "__main__":
    print(f"🚀 启动演示程序...")
    print(f"CUDA 版本: {paddle.device.cuda.get_device_name()}")
    print(f"CUDA_VISIBLE_DEVICES: {os.environ.get('CUDA_VISIBLE_DEVICES')}")
    available_gpus = detect_available_gpus()
    monitor_gpu_memory(available_gpus)

    input_path = "./sample_data/300674-母公司现金流量表-扫描.png"

    # pipeline_path = "./PP-StructureV3-zhch.yaml"
    pipeline_path = "PP-StructureV3"
    pipeline_name = Path(pipeline_path).stem
    output_path = Path(f"./sample_data/single_pipeline_output/{pipeline_name}/")

    pipeline = create_pipeline(pipeline=pipeline_path, device="gpu:0")

    # For Image
    output = pipeline.predict(
        input=input_path,
        # device="gpu",  # 或者 "gpu" 如果你有 GPU 支持
        use_doc_orientation_classify=True, # 开启文档方向分类
        use_doc_unwarping=False, # 开启文档去畸变, 效果不佳
        layout_detection_model_name=None, # 如果要禁用版面分析，可以这样设置，或者依赖其默认行为
        use_seal_recognition=True,         # 跳过印章识别
        use_chart_recognition=True,         # 跳过图表识别
        use_table_recognition=True,        # 开启表格识别
    )

    # 可视化结果并保存 json 结果
    for res in output:
        res.print() 
        # res.save_to_json(save_path="sample_data/output") 
        # res.save_to_markdown(save_path="sample_data/output") 
        res.save_all(save_path=output_path.as_posix())  # 保存所有结果到指定路径