yohann84L

def reduce_mem_usage(props):
    """
    Code seen on https://www.kaggle.com/jesucristo/fraud-complete-eda
    to reduce memory usage of a dataframe by using the write
    dtype for each variable.
    
    /!\ Be careful with it's usage, dat could be broken after /!\
    
    NaN value are replaced by -1
    """

yohann84L

def get_pdf_without_text(pdf: fitz.Document) -> fitz.Document:
    for i in range(0, pdf.pageCount):
        page = pdf[i]
        for xref in page._getContents():
            stream = pdf._getXrefStream(xref).replace(b'Tm', b'Tm\n3 Tr')
            pdf._updateStream(xref, stream)
    return pdf

yohann84L

## GridMask augmentation for imgaug
##
## Code based on this kernel https://www.kaggle.com/shivyshiv/efficientnet-gridmask-training-pytorch



from imgaug.augmenters import meta
from imgaug import parameters as iap
import imgaug.augmenters as iaa

yohann84L

import random

import numpy as np
from PIL import Image, ImageEnhance, ImageOps


class ImageNetPolicy(object):
    def __init__(self, fillcolor=(128, 128, 128)):
        self.policies = [
            SubPolicy(0.4, "posterize", 8, 0.6, "rotate", 9, fillcolor),