rvl_cdip_multi / test_loader.py
jordyvl's picture
try to run for test set image conversion
49508ed
import os
from io import BytesIO
from tqdm import tqdm
from typing import Callable, Dict, List, Literal, NamedTuple, Optional, Tuple, Union
from datasets import load_dataset, load_dataset_builder, load_from_disk
from PIL import Image
from codetiming import Timer
import pdf2image
import PyPDF2
import pikepdf
import fitz as PyMuPDF
#https://gist.github.com/gstorer/f6a9f1dfe41e8e64dcf58d07afa9ab2a
# if os.getlogin() == "jordy": #local testing
# DATASET = "/home/jordy/code/opensource/Beyond-Document-Page-Classification/src/tests/rvl_cdip_multi/rvl_cdip_multi.py"
# stepwise = True
DATASET = "jordyvl/rvl_cdip_multi"
stepwise = False
testds = load_dataset("jordyvl/rvl_cdip_multi", cache_dir="/mnt/lerna/data/HFcache", split="test")
# time how long each way to pdf2image lasts
def batched_conversion(pdf_file):
info = pdf2image.pdfinfo_from_path(pdf_file, userpw=None, poppler_path=None)
maxPages = info["Pages"]
logger.info(f"{pdf_file} has {str(maxPages)} pages")
images = []
for page in range(1, maxPages + 1, 10):
images.extend(
pdf2image.convert_from_path(
pdf_file, dpi=200, first_page=page, last_page=min(page + 10 - 1, maxPages)
)
)
return images
def batch_get_images(streams, wrapper):
return {"images": [get_images(pdf_stream, wrapper=wrapper) for pdf_stream in streams]}
def batch_get_pages(streams, wrapper):
return {"pages": [get_pages(pdf_stream, wrapper=wrapper) for pdf_stream in streams]}
def get_pages(pdf_stream, wrapper="pike"):
if wrapper == "pdf2image":
return pdf2image.pdfinfo_from_bytes(pdf_stream, userpw=None, poppler_path=None)["Pages"]
pdf_file = BytesIO(pdf_stream)
if wrapper == "pike":
return len(pikepdf.Pdf.open(pdf_file).pages)
if wrapper == "PyPDF2":
return len(PyPDF2.PdfReader(pdf_file).pages)
if wrapper == "pymupdf":
return len(PyMuPDF.open(stream=pdf_stream, filetype="pdf"))
def pymupdf_image_extraction(data: bytes):
images = []
with PyMuPDF.open(stream=data, filetype="pdf") as pdf_file:
for page_index in range(len(pdf_file)):
page = pdf_file[page_index]
for image_index, img in enumerate(page.get_images(), start=1):
xref = img[0]
base_image = pdf_file.extract_image(xref)
image_bytes = base_image["image"]
image_ext = base_image["ext"]
im = Image.open(BytesIO(image_bytes))
images.append(
im
#(f"image{page_index+1}_{image_index}.{image_ext}", image_bytes)
)
return images
def pypdf_image_extraction(data: bytes): # -> List[Tuple[str, bytes]]
images = []
try:
reader = PyPDF2.PdfReader(BytesIO(data))
for page in reader.pages:
for image in page.images:
im = Image.open(BytesIO(image.data))
images.append(im)
#images.append(Image.frombytes("RGB",(image.width,image.height),image.data)) #(image.name, #width,height
except Exception as exc:
print(f"PyPDF2 Image extraction failure: {exc}")
return images
def pike_image_extraction(data: bytes):
images = []
reader = pikepdf.Pdf.open(BytesIO(data))
for page in reader.pages:
for raw_name, raw_image in page.images.items():
image = pikepdf.PdfImage(raw_image).as_pil_image()
images.append(image)
return images
def get_images(pdf_stream, wrapper="pike"):
if wrapper == "pike":
return pike_image_extraction(pdf_stream)
if wrapper == "pdf2image":
return pdf2image.convert_from_bytes(pdf_stream)
if wrapper == "pymupdf":
return pymupdf_image_extraction(pdf_stream)
if wrapper == "PyPDF2":
return pypdf_image_extraction(pdf_stream)
#use same tool for page and images
subset = testds.select(list(range(0, 100)))
images_per_binding = {}
for binding in ["pike", "pdf2image", "PyPDF2", "pymupdf"]:
with Timer(name=f"{binding}", text=binding + " Elapsed time: {:.4f} seconds"):
func = lambda batch: batch_get_images(batch["file"], wrapper=binding)
images_per_binding[binding] = subset.map(func, batched=True, keep_in_memory=False)
"""
Image.open(BytesIO(images_per_binding["pymupdf"]["images"][0][0]["bytes"])).show()
Image.open(BytesIO(images_per_binding["PyPDF2"]["images"][0][0]["bytes"])).show()
Image.open(BytesIO(images_per_binding["pike"]["images"][0][0]["bytes"])).show()
Image.open(BytesIO(images_per_binding["pdf2image"]["images"][0][0]["bytes"])).show()
"""
# whut?
# now they are PIL: Image.open(BytesIO(images_per_binding["pymupdf"]["images"][0][0]["bytes"]))
import pdb; pdb.set_trace() # breakpoint 01d24a47 //
if stepwise:
ds = load_dataset(DATASET, cache_dir="/mnt/lerna/data/HFcache")
import pdb
pdb.set_trace() # breakpoint 1898e7e0 //
builder = load_dataset_builder(DATASET)
ds = builder.download_and_prepare()
ds = builder.as_dataset(split="test")
# count pages
counter = []
for i, d in tqdm(enumerate(ds["test"])):
counter.append(len(d["file"]))
print(sum(counter) / len(counter))
"""
pdfinfo ylhw0010.pdf
Creator: CaptureSDK ENGine - www.scansoft.com
Producer: Scansoft Capture Development System V12.0; modified using iText 2.1.7 by 1T3XT
CreationDate: Sat Jan 1 01:00:00 2005 CET
ModDate: Sat Dec 29 08:04:44 2018 CET
Tagged: no
UserProperties: no
Suspects: no
Form: none
JavaScript: no
Pages: 1
Encrypted: no
Page size: 576 x 756 pts
Page rot: 0
File size: 24443 bytes
Optimized: yes
PDF version: 1.3
jordy@jordy-OMEN:~/Downl
"""