PDF转markdown工具：magic-pdf

1. magic-pdf 环境安装

conda create -n MinerU python=3.10
conda activate MinerU
pip install boto3>=1.28.43 -i https://pypi.tuna.tsinghua.edu.cn/simple/
pip install magic-pdf[full]==0.7.0b1 --extra-index-url https://wheels.myhloli.com  -i https://pypi.tuna.tsinghua.edu.cn/simple/

2. 权重下载

sudo apt-get install git-lfs
git clone https://github.com/opendatalab/MinerU.git
cd MinerU/
git lfs install
git lfs clone https://huggingface.co/wanderkid/PDF-Extract-Kit

或者

pip install modelscope

python 复制代码

# Use the following Python code to download the model using the ModelScope SDK:
from modelscope import snapshot_download
model_dir = snapshot_download('wanderkid/PDF-Extract-Kit')

3. 修改配置

修改

复制代码

magic-pdf.template.json 中models-dir修改为模型的下载路径

python 复制代码

{
    "bucket_info":{
        "bucket-name-1":["ak", "sk", "endpoint"],
        "bucket-name-2":["ak", "sk", "endpoint"]
    },
    "models-dir":"/home/adam/work/MinerU/PDF-Extract-Kit/models",
    "device-mode":"cpu",
    "table-config": {
        "is_table_recog_enable": false,
        "max_time": 400
    }
}

将magic-pdf.template.json文件修改为magic-pdf.json放在系统目录，不同的系统默认目录不同，

Windows ： C:\Users\YourUsername,

Linux ： /home/YourUsername

macOS ： /Users/YourUsername

4. 使用参数

python 复制代码

magic-pdf --help
Usage: magic-pdf [OPTIONS]

Options:
  -v, --version                display the version and exit
  -p, --path PATH              local pdf filepath or directory  [required]
  -o, --output-dir TEXT        output local directory
  -m, --method [ocr|txt|auto]  the method for parsing pdf.  
                               ocr: using ocr technique to extract information from pdf,
                               txt: suitable for the text-based pdf only and outperform ocr,
                               auto: automatically choose the best method for parsing pdf
                                  from ocr and txt.
                               without method specified, auto will be used by default. 
  --help                       Show this message and exit.


## show version
magic-pdf -v

## command line example
magic-pdf -p {some_pdf} -o {some_output_dir} -m auto

{some_pdf}可以是单个 PDF 文件，也可以是包含多个 PDF 的目录。结果将保存在目录中。输出文件列表如下：{some_output_dir}

python 复制代码

├── some_pdf.md                 # markdown file
├── images                      # directory for storing images
├── layout.pdf                  # layout diagram
├── middle.json                 # MinerU intermediate processing result
├── model.json                  # model inference result
├── origin.pdf                  # original PDF file
└── spans.pdf                   # smallest granularity bbox position information diagram

5.测试

magic-pdf -p GenZ-LLM.pdf -o ./res/ -m auto

结果：

测试使用cpu执行，内存16g，3页pdf解析大概2分钟, 页数过多会崩掉。有些公式好像解析的不太对，整体可用。

具体log:

2024-08-13 15:53:44.149 | INFO     | magic_pdf.libs.pdf_check:detect_invalid_chars:57 - cid_count: 0, text_len: 14962, cid_chars_radio: 0.0
INFO:datasets:PyTorch version 2.3.1 available.
2024-08-13 15:53:53.048 | INFO     | magic_pdf.model.pdf_extract_kit:__init__:111 - DocAnalysis init, this may take some times. apply_layout: True, apply_formula: True, apply_ocr: False, apply_table: False
2024-08-13 15:53:53.048 | INFO     | magic_pdf.model.pdf_extract_kit:__init__:119 - using device: cpu
2024-08-13 15:53:53.048 | INFO     | magic_pdf.model.pdf_extract_kit:__init__:121 - using models_dir: /home/long/work/MinerU/PDF-Extract-Kit/models
CustomVisionEncoderDecoderModel init
CustomMBartForCausalLM init
CustomMBartDecoder init
[08/13 15:54:06 detectron2]: Rank of current process: 0. World size: 1
[08/13 15:54:07 detectron2]: Environment info:
-------------------------------  --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
sys.platform                     linux
Python                           3.10.14 (main, May  6 2024, 19:42:50) [GCC 11.2.0]
numpy                            1.26.4
detectron2                       0.6 @/home/long/anaconda3/envs/MinerU/lib/python3.10/site-packages/detectron2
detectron2._C                    not built correctly: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.32' not found (required by /home/long/anaconda3/envs/MinerU/lib/python3.10/site-packages/detectron2/_C.cpython-310-x86_64-linux-gnu.so)
Compiler ($CXX)                  c++ (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0
DETECTRON2_ENV_MODULE            <not set>
PyTorch                          2.3.1+cu121 @/home/long/anaconda3/envs/MinerU/lib/python3.10/site-packages/torch
PyTorch debug build              False
torch._C._GLIBCXX_USE_CXX11_ABI  False
GPU available                    No: torch.cuda.is_available() == False
Pillow                           10.4.0
torchvision                      0.18.1+cu121 @/home/long/anaconda3/envs/MinerU/lib/python3.10/site-packages/torchvision
fvcore                           0.1.5.post20221221
iopath                           0.1.9
cv2                              4.6.0
-------------------------------  --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
PyTorch built with:
  - GCC 9.3
  - C++ Version: 201703
  - Intel(R) oneAPI Math Kernel Library Version 2022.2-Product Build 20220804 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v3.3.6 (Git Hash 86e6af5974177e513fd3fee58425e1063e7f1361)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=12.1, CUDNN_VERSION=8.9.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -D_GLIBCXX_USE_CXX11_ABI=0 -fabi-version=11 -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=pedantic -Wno-error=old-style-cast -Wno-missing-braces -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=2.3.1, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, 

[08/13 15:54:07 detectron2]: Command line arguments: {'config_file': '/home/long/anaconda3/envs/MinerU/lib/python3.10/site-packages/magic_pdf/resources/model_config/layoutlmv3/layoutlmv3_base_inference.yaml', 'resume': False, 'eval_only': False, 'num_gpus': 1, 'num_machines': 1, 'machine_rank': 0, 'dist_url': 'tcp://127.0.0.1:57823', 'opts': ['MODEL.WEIGHTS', '/home/long/work/MinerU/PDF-Extract-Kit/models/Layout/model_final.pth']}
[08/13 15:54:07 detectron2]: Contents of args.config_file=/home/long/anaconda3/envs/MinerU/lib/python3.10/site-packages/magic_pdf/resources/model_config/layoutlmv3/layoutlmv3_base_inference.yaml:
AUG:
  DETR: true
CACHE_DIR: ~/cache/huggingface
CUDNN_BENCHMARK: false
DATALOADER:
  ASPECT_RATIO_GROUPING: true
  FILTER_EMPTY_ANNOTATIONS: false
  NUM_WORKERS: 4
  REPEAT_THRESHOLD: 0.0
  SAMPLER_TRAIN: TrainingSampler
DATASETS:
  PRECOMPUTED_PROPOSAL_TOPK_TEST: 1000
  PRECOMPUTED_PROPOSAL_TOPK_TRAIN: 2000
  PROPOSAL_FILES_TEST: []
  PROPOSAL_FILES_TRAIN: []
  TEST:
  - scihub_train
  TRAIN:
  - scihub_train
GLOBAL:
  HACK: 1.0
ICDAR_DATA_DIR_TEST: ''
ICDAR_DATA_DIR_TRAIN: ''
INPUT:
  CROP:
    ENABLED: true
    SIZE:
    - 384
    - 600
    TYPE: absolute_range
  FORMAT: RGB
  MASK_FORMAT: polygon
  MAX_SIZE_TEST: 1333
  MAX_SIZE_TRAIN: 1333
  MIN_SIZE_TEST: 800
  MIN_SIZE_TRAIN:
  - 480
  - 512
  - 544
  - 576
  - 608
  - 640
  - 672
  - 704
  - 736
  - 768
  - 800
  MIN_SIZE_TRAIN_SAMPLING: choice
  RANDOM_FLIP: horizontal
MODEL:
  ANCHOR_GENERATOR:
    ANGLES:
    - - -90
      - 0
      - 90
    ASPECT_RATIOS:
    - - 0.5
      - 1.0
      - 2.0
    NAME: DefaultAnchorGenerator
    OFFSET: 0.0
    SIZES:
    - - 32
    - - 64
    - - 128
    - - 256
    - - 512
  BACKBONE:
    FREEZE_AT: 2
    NAME: build_vit_fpn_backbone
  CONFIG_PATH: ''
  DEVICE: cuda
  FPN:
    FUSE_TYPE: sum
    IN_FEATURES:
    - layer3
    - layer5
    - layer7
    - layer11
    NORM: ''
    OUT_CHANNELS: 256
  IMAGE_ONLY: true
  KEYPOINT_ON: false
  LOAD_PROPOSALS: false
  MASK_ON: true
  META_ARCHITECTURE: VLGeneralizedRCNN
  PANOPTIC_FPN:
    COMBINE:
      ENABLED: true
      INSTANCES_CONFIDENCE_THRESH: 0.5
      OVERLAP_THRESH: 0.5
      STUFF_AREA_LIMIT: 4096
    INSTANCE_LOSS_WEIGHT: 1.0
  PIXEL_MEAN:
  - 127.5
  - 127.5
  - 127.5
  PIXEL_STD:
  - 127.5
  - 127.5
  - 127.5
  PROPOSAL_GENERATOR:
    MIN_SIZE: 0
    NAME: RPN
  RESNETS:
    DEFORM_MODULATED: false
    DEFORM_NUM_GROUPS: 1
    DEFORM_ON_PER_STAGE:
    - false
    - false
    - false
    - false
    DEPTH: 50
    NORM: FrozenBN
    NUM_GROUPS: 1
    OUT_FEATURES:
    - res4
    RES2_OUT_CHANNELS: 256
    RES5_DILATION: 1
    STEM_OUT_CHANNELS: 64
    STRIDE_IN_1X1: true
    WIDTH_PER_GROUP: 64
  RETINANET:
    BBOX_REG_LOSS_TYPE: smooth_l1
    BBOX_REG_WEIGHTS:
    - 1.0
    - 1.0
    - 1.0
    - 1.0
    FOCAL_LOSS_ALPHA: 0.25
    FOCAL_LOSS_GAMMA: 2.0
    IN_FEATURES:
    - p3
    - p4
    - p5
    - p6
    - p7
    IOU_LABELS:
    - 0
    - -1
    - 1
    IOU_THRESHOLDS:
    - 0.4
    - 0.5
    NMS_THRESH_TEST: 0.5
    NORM: ''
    NUM_CLASSES: 10
    NUM_CONVS: 4
    PRIOR_PROB: 0.01
    SCORE_THRESH_TEST: 0.05
    SMOOTH_L1_LOSS_BETA: 0.1
    TOPK_CANDIDATES_TEST: 1000
  ROI_BOX_CASCADE_HEAD:
    BBOX_REG_WEIGHTS:
    - - 10.0
      - 10.0
      - 5.0
      - 5.0
    - - 20.0
      - 20.0
      - 10.0
      - 10.0
    - - 30.0
      - 30.0
      - 15.0
      - 15.0
    IOUS:
    - 0.5
    - 0.6
    - 0.7
  ROI_BOX_HEAD:
    BBOX_REG_LOSS_TYPE: smooth_l1
    BBOX_REG_LOSS_WEIGHT: 1.0
    BBOX_REG_WEIGHTS:
    - 10.0
    - 10.0
    - 5.0
    - 5.0
    CLS_AGNOSTIC_BBOX_REG: true
    CONV_DIM: 256
    FC_DIM: 1024
    NAME: FastRCNNConvFCHead
    NORM: ''
    NUM_CONV: 0
    NUM_FC: 2
    POOLER_RESOLUTION: 7
    POOLER_SAMPLING_RATIO: 0
    POOLER_TYPE: ROIAlignV2
    SMOOTH_L1_BETA: 0.0
    TRAIN_ON_PRED_BOXES: false
  ROI_HEADS:
    BATCH_SIZE_PER_IMAGE: 512
    IN_FEATURES:
    - p2
    - p3
    - p4
    - p5
    IOU_LABELS:
    - 0
    - 1
    IOU_THRESHOLDS:
    - 0.5
    NAME: CascadeROIHeads
    NMS_THRESH_TEST: 0.5
    NUM_CLASSES: 10
    POSITIVE_FRACTION: 0.25
    PROPOSAL_APPEND_GT: true
    SCORE_THRESH_TEST: 0.05
  ROI_KEYPOINT_HEAD:
    CONV_DIMS:
    - 512
    - 512
    - 512
    - 512
    - 512
    - 512
    - 512
    - 512
    LOSS_WEIGHT: 1.0
    MIN_KEYPOINTS_PER_IMAGE: 1
    NAME: KRCNNConvDeconvUpsampleHead
    NORMALIZE_LOSS_BY_VISIBLE_KEYPOINTS: true
    NUM_KEYPOINTS: 17
    POOLER_RESOLUTION: 14
    POOLER_SAMPLING_RATIO: 0
    POOLER_TYPE: ROIAlignV2
  ROI_MASK_HEAD:
    CLS_AGNOSTIC_MASK: false
    CONV_DIM: 256
    NAME: MaskRCNNConvUpsampleHead
    NORM: ''
    NUM_CONV: 4
    POOLER_RESOLUTION: 14
    POOLER_SAMPLING_RATIO: 0
    POOLER_TYPE: ROIAlignV2
  RPN:
    BATCH_SIZE_PER_IMAGE: 256
    BBOX_REG_LOSS_TYPE: smooth_l1
    BBOX_REG_LOSS_WEIGHT: 1.0
    BBOX_REG_WEIGHTS:
    - 1.0
    - 1.0
    - 1.0
    - 1.0
    BOUNDARY_THRESH: -1
    CONV_DIMS:
    - -1
    HEAD_NAME: StandardRPNHead
    IN_FEATURES:
    - p2
    - p3
    - p4
    - p5
    - p6
    IOU_LABELS:
    - 0
    - -1
    - 1
    IOU_THRESHOLDS:
    - 0.3
    - 0.7
    LOSS_WEIGHT: 1.0
    NMS_THRESH: 0.7
    POSITIVE_FRACTION: 0.5
    POST_NMS_TOPK_TEST: 1000
    POST_NMS_TOPK_TRAIN: 2000
    PRE_NMS_TOPK_TEST: 1000
    PRE_NMS_TOPK_TRAIN: 2000
    SMOOTH_L1_BETA: 0.0
  SEM_SEG_HEAD:
    COMMON_STRIDE: 4
    CONVS_DIM: 128
    IGNORE_VALUE: 255
    IN_FEATURES:
    - p2
    - p3
    - p4
    - p5
    LOSS_WEIGHT: 1.0
    NAME: SemSegFPNHead
    NORM: GN
    NUM_CLASSES: 10
  VIT:
    DROP_PATH: 0.1
    IMG_SIZE:
    - 224
    - 224
    NAME: layoutlmv3_base
    OUT_FEATURES:
    - layer3
    - layer5
    - layer7
    - layer11
    POS_TYPE: abs
  WEIGHTS: 
OUTPUT_DIR: 
SCIHUB_DATA_DIR_TRAIN: ~/publaynet/layout_scihub/train
SEED: 42
SOLVER:
  AMP:
    ENABLED: true
  BACKBONE_MULTIPLIER: 1.0
  BASE_LR: 0.0002
  BIAS_LR_FACTOR: 1.0
  CHECKPOINT_PERIOD: 2000
  CLIP_GRADIENTS:
    CLIP_TYPE: full_model
    CLIP_VALUE: 1.0
    ENABLED: true
    NORM_TYPE: 2.0
  GAMMA: 0.1
  GRADIENT_ACCUMULATION_STEPS: 1
  IMS_PER_BATCH: 32
  LR_SCHEDULER_NAME: WarmupCosineLR
  MAX_ITER: 20000
  MOMENTUM: 0.9
  NESTEROV: false
  OPTIMIZER: longW
  REFERENCE_WORLD_SIZE: 0
  STEPS:
  - 10000
  WARMUP_FACTOR: 0.01
  WARMUP_ITERS: 333
  WARMUP_METHOD: linear
  WEIGHT_DECAY: 0.05
  WEIGHT_DECAY_BIAS: null
  WEIGHT_DECAY_NORM: 0.0
TEST:
  AUG:
    ENABLED: false
    FLIP: true
    MAX_SIZE: 4000
    MIN_SIZES:
    - 400
    - 500
    - 600
    - 700
    - 800
    - 900
    - 1000
    - 1100
    - 1200
  DETECTIONS_PER_IMAGE: 100
  EVAL_PERIOD: 1000
  EXPECTED_RESULTS: []
  KEYPOINT_OKS_SIGMAS: []
  PRECISE_BN:
    ENABLED: false
    NUM_ITER: 200
VERSION: 2
VIS_PERIOD: 0

[08/13 15:54:08 d2.checkpoint.detection_checkpoint]: [DetectionCheckpointer] Loading from /home/long/work/MinerU/PDF-Extract-Kit/models/Layout/model_final.pth ...
[08/13 15:54:08 fvcore.common.checkpoint]: [Checkpointer] Loading from /home/long/work/MinerU/PDF-Extract-Kit/models/Layout/model_final.pth ...
2024-08-13 15:54:09.334 | INFO     | magic_pdf.model.pdf_extract_kit:__init__:148 - DocAnalysis init done!
2024-08-13 15:54:09.336 | INFO     | magic_pdf.model.doc_analyze_by_custom_model:custom_model_init:98 - model init cost: 25.18623661994934
2024-08-13 15:54:18.411 | INFO     | magic_pdf.model.pdf_extract_kit:__call__:159 - layout detection cost: 8.96

0: 1888x1472 2 embeddings, 3839.2ms
Speed: 28.6ms preprocess, 3839.2ms inference, 0.9ms postprocess per image at shape (1, 3, 1888, 1472)
2024-08-13 15:54:25.349 | INFO     | magic_pdf.model.pdf_extract_kit:__call__:189 - formula nums: 2, mfr time: 1.24
2024-08-13 15:54:34.577 | INFO     | magic_pdf.model.pdf_extract_kit:__call__:159 - layout detection cost: 9.22

0: 1888x1472 25 embeddings, 4120.5ms
Speed: 15.3ms preprocess, 4120.5ms inference, 1.0ms postprocess per image at shape (1, 3, 1888, 1472)
2024-08-13 15:54:49.462 | INFO     | magic_pdf.model.pdf_extract_kit:__call__:189 - formula nums: 25, mfr time: 10.67
2024-08-13 15:54:59.903 | INFO     | magic_pdf.model.pdf_extract_kit:__call__:159 - layout detection cost: 10.44

0: 1888x1472 18 embeddings, 4241.8ms
Speed: 20.1ms preprocess, 4241.8ms inference, 0.9ms postprocess per image at shape (1, 3, 1888, 1472)
2024-08-13 15:55:12.180 | INFO     | magic_pdf.model.pdf_extract_kit:__call__:189 - formula nums: 18, mfr time: 7.93
2024-08-13 15:55:12.184 | INFO     | magic_pdf.model.doc_analyze_by_custom_model:doc_analyze:124 - doc analyze cost: 62.73242211341858
2024-08-13 15:55:12.233 | INFO     | magic_pdf.pdf_parse_union_core:pdf_parse_union:221 - page_id: 0, last_page_cost_time: 0.0
2024-08-13 15:55:12.305 | INFO     | magic_pdf.pdf_parse_union_core:pdf_parse_union:221 - page_id: 1, last_page_cost_time: 0.07
2024-08-13 15:55:12.364 | INFO     | magic_pdf.pdf_parse_union_core:pdf_parse_union:221 - page_id: 2, last_page_cost_time: 0.06
2024-08-13 15:55:12.743 | INFO     | magic_pdf.para.para_split_v2:__detect_list_lines:140 - 发现了列表，列表行数：[(8, 9)]， [[8, 9]]
2024-08-13 15:55:12.744 | INFO     | magic_pdf.para.para_split_v2:__detect_list_lines:153 - 列表行的第8到第9行是列表
2024-08-13 15:55:12.750 | INFO     | magic_pdf.para.para_split_v2:__detect_list_lines:140 - 发现了列表，列表行数：[(19, 20)]， [[19]]
2024-08-13 15:55:12.750 | INFO     | magic_pdf.para.para_split_v2:__detect_list_lines:153 - 列表行的第19到第20行是列表
2024-08-13 15:55:12.755 | INFO     | magic_pdf.para.para_split_v2:para_split:764 - 连接了第0页和第1页的段落
2024-08-13 15:55:13.239 | INFO     | magic_pdf.pipe.UNIPipe:pipe_mk_markdown:48 - uni_pipe mk mm_markdown finished
2024-08-13 15:55:13.278 | INFO     | magic_pdf.pipe.UNIPipe:pipe_mk_uni_format:43 - uni_pipe mk content list finished
2024-08-13 15:55:13.278 | INFO     | magic_pdf.tools.common:do_parse:119 - local output dir is ./res/GenZ-LLM-Analyzer/auto