Hugging Face's logo

Datasets:
TrainingDataPro
/
generated-usa-passeports-dataset

Tasks:
Image-to-Image
Languages:
English
License:
Dataset card Files Community
generated-usa-passeports-dataset / generated-usa-passeports-dataset.py
Vadzim Kashko
fix: image representation as np.ndarray
3cb14b5
raw history blame contribute delete
No virus
5.66 kB
import datasets
import PIL.Image
import PIL.ImageOps
import numpy as np
_CITATION = """\
@InProceedings{huggingface:dataset,
title = {generated-usa-passeports-dataset},
author = {TrainingDataPro},
year = {2023}
}
"""
_DESCRIPTION = """\
Data generation in machine learning involves creating or manipulating data
to train and evaluate machine learning models. The purpose of data generation
is to provide diverse and representative examples that cover a wide range of
scenarios, ensuring the model's robustness and generalization.
Data augmentation techniques involve applying various transformations to
existing data samples to create new ones. These transformations include:
random rotations, translations, scaling, flips, and more. Augmentation helps
in increasing the dataset size, introducing natural variations, and improving
model performance by making it more invariant to specific transformations.
The dataset contains **GENERATED** USA passports, which are replicas of
official passports but with randomly generated details, such as name, date of
birth etc. The primary intention of generating these fake passports is to
demonstrate the structure and content of a typical passport document and to
train the neural network to identify this type of document.
Generated passports can assist in conducting research without accessing or
compromising real user data that is often sensitive and subject to privacy
regulations. Synthetic data generation allows researchers to develop and
refine models using simulated passport data without risking privacy leaks.
"""
_NAME = 'generated-usa-passeports-dataset'
_HOMEPAGE = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}"
_LICENSE = "cc-by-nc-nd-4.0"
_DATA = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}/resolve/main/data/"
def exif_transpose(img):
if not img:
return img
exif_orientation_tag = 274
# Check for EXIF data (only present on some files)
if hasattr(img, "_getexif") and isinstance(
img._getexif(), dict) and exif_orientation_tag in img._getexif():
exif_data = img._getexif()
orientation = exif_data[exif_orientation_tag]
# Handle EXIF Orientation
if orientation == 1:
# Normal image - nothing to do!
pass
elif orientation == 2:
# Mirrored left to right
img = img.transpose(PIL.Image.FLIP_LEFT_RIGHT)
elif orientation == 3:
# Rotated 180 degrees
img = img.rotate(180)
elif orientation == 4:
# Mirrored top to bottom
img = img.rotate(180).transpose(PIL.Image.FLIP_LEFT_RIGHT)
elif orientation == 5:
# Mirrored along top-left diagonal
img = img.rotate(-90,
expand=True).transpose(PIL.Image.FLIP_LEFT_RIGHT)
elif orientation == 6:
# Rotated 90 degrees
img = img.rotate(-90, expand=True)
elif orientation == 7:
# Mirrored along top-right diagonal
img = img.rotate(90,
expand=True).transpose(PIL.Image.FLIP_LEFT_RIGHT)
elif orientation == 8:
# Rotated 270 degrees
img = img.rotate(90, expand=True)
return img
def load_image_file(file, mode='RGB'):
# Load the image with PIL
img = PIL.Image.open(file)
if hasattr(PIL.ImageOps, 'exif_transpose'):
# Very recent versions of PIL can do exit transpose internally
img = PIL.ImageOps.exif_transpose(img)
else:
# Otherwise, do the exif transpose ourselves
img = exif_transpose(img)
img = img.convert(mode)
return np.array(img)
class GeneratedUsaPasseportsDataset(datasets.GeneratorBasedBuilder):
def _info(self):
return datasets.DatasetInfo(
description=_DESCRIPTION,
features=datasets.Features({
'original': datasets.Image(),
'us_pass_augmentated_1': datasets.Image(),
'us_pass_augmentated_2': datasets.Image(),
'us_pass_augmentated_3': datasets.Image()
}),
supervised_keys=None,
homepage=_HOMEPAGE,
citation=_CITATION,
license=_LICENSE)
def _split_generators(self, dl_manager):
original = dl_manager.download_and_extract(f"{_DATA}original.zip")
augmentation = dl_manager.download_and_extract(
f"{_DATA}augmentation.zip")
annotations = dl_manager.download(f"{_DATA}{_NAME}.csv")
original = dl_manager.iter_files(original)
augmentation = dl_manager.iter_files(augmentation)
return [
datasets.SplitGenerator(name=datasets.Split.TRAIN,
gen_kwargs={
"original": original,
'augmentation': augmentation,
'annotations': annotations
}),
]
def _generate_examples(self, original, augmentation, annotations):
original = list(original)
augmentation = list(augmentation)
augmentation = [
augmentation[i:i + 3] for i in range(0, len(augmentation), 3)
]
for idx, (org, aug) in enumerate(zip(original, augmentation)):
yield idx, {
'original': load_image_file(org),
'us_pass_augmentated_1': load_image_file(aug[0]),
'us_pass_augmentated_2': load_image_file(aug[1]),
'us_pass_augmentated_3': load_image_file(aug[2])
}