def extract(self, pdf_path, output_directory):
"""Return a ``FigureExtraction`` instance for ``pdf_path``.
Extract the figures and additional information from the PDF at
``pdf_path``, saving the results to disk in ``output_directory``
and returning the corresponding ``FigureExtraction`` instance.
Parameters
----------
pdf_path : str
The path to the PDF.
output_directory : str
The directory in which to save the results from extraction.
Returns
-------
FigureExtraction
A ``FigureExtraction`` instance for the PDF at ``pdf_path``.
"""
figure_extraction = FigureExtraction(pdf_path=pdf_path,
parent_directory=output_directory)
# create the extraction results directory
os.makedirs(figure_extraction.paths['BASE'])
# copy the PDF into the extraction results directory
shutil.copy(pdf_path, figure_extraction.paths['PDF_PATH'])
pdf_renderer = settings_utils.import_setting(
settings.DEEPFIGURES_PDF_RENDERER)()
# render the PDF into low-res images
figure_extraction.low_res_rendering_paths = \
pdf_renderer.render(
pdf_path=figure_extraction.paths['PDF_PATH'],
output_dir=figure_extraction.paths['BASE'],
dpi=settings.DEFAULT_INFERENCE_DPI)
# render the PDF into hi-res images
figure_extraction.hi_res_rendering_paths = \
pdf_renderer.render(
pdf_path=figure_extraction.paths['PDF_PATH'],
output_dir=figure_extraction.paths['BASE'],
dpi=settings.DEFAULT_CROPPED_IMG_DPI)
# extract captions from PDF using pdffigures2
figure_extraction.pdffigures_output_path = \
pdffigures_wrapper.pdffigures_extractor.extract(
pdf_path=figure_extraction.paths['PDF_PATH'],
output_dir=figure_extraction.paths['BASE'])
# run deepfigures / neural networks on the PDF images
figure_extraction.deepfigures_json_path = \
detection.extract_figures_json(
pdf_path=figure_extraction.paths['PDF_PATH'],
page_image_paths=figure_extraction.low_res_rendering_paths,
pdffigures_output=figure_extraction.pdffigures_output_path,
output_directory=figure_extraction.paths['BASE'])
return figure_extraction
def convert_pdf_paths_to_images(input_dir,
output_dir,
temp,
cpu_count=cpu_count(),
max_pages=500):
pdf_renderer = settings_utils.import_setting(
settings.DEEPFIGURES_PDF_RENDERER)()
pdf_paths = glob.glob(os.path.join(input_dir, '*.pdf'))
logger.info(
"Total paths obtained: {path_count}".format(path_count=len(pdf_paths)))
with Pool(cpu_count) as pool:
pool.map(
partial(convert_pdf_to_images,
_output_dir=output_dir,
_pdf_renderer=pdf_renderer,
_temp_dir=temp,
max_pages=max_pages), pdf_paths)
import numpy as np
import tensorflow as tf
from deepfigures import settings
from deepfigures.extraction.datamodels import (BoxClass, Figure,
PdfDetectionResult, CaptionOnly)
from deepfigures.extraction import (figure_utils, pdffigures_wrapper)
from deepfigures.extraction.pdffigures_wrapper import pdffigures_extractor
from deepfigures.utils import (file_util, image_util, settings_utils)
import train
from vendor.tensorboxresnet.tensorboxresnet.utils import train_utils
CAPTION_CHANNEL_BACKGROUND = 255
CAPTION_CHANNEL_MASK = 0
pdf_renderer = settings_utils.import_setting(
settings.DEEPFIGURES_PDF_RENDERER)()
class TensorboxCaptionmaskDetector(object):
"""Interface for using the neural network model to detect figures.
Instantiating this class creates a tensorflow session object as the
self.sess attribute. When done using the instance, remember to close
the session; however, do not open and close sessions every time you
extract a figure because the added overhead will very negatively
affect performance.
"""
def __init__(
self,
save_dir,
iteration,