license_plates / license_plates.py
vkashko's picture
feat: add bbox_id
0cff0b4
raw
history blame
4.27 kB
import datasets
import pandas as pd
_CITATION = """\
@InProceedings{huggingface:dataset,
title = {license_plates},
author = {TrainingDataPro},
year = {2023}
}
"""
_DESCRIPTION = """\
Over 1.2 million annotated license plates from vehicles around the world.
This dataset is tailored for License Plate Recognition tasks and includes
images from both YouTube and PlatesMania.
Annotation details are provided in the About section below.
"""
_NAME = 'license_plates'
_HOMEPAGE = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}"
_LICENSE = ""
_DATA = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}/resolve/main/data/"
class LicensePlates(datasets.GeneratorBasedBuilder):
BUILDER_CONFIGS = [
datasets.BuilderConfig(name="Brazil_youtube"),
datasets.BuilderConfig(name="Estonia_platesmania"),
datasets.BuilderConfig(name="Finland_platesmania"),
datasets.BuilderConfig(name="Kazakhstan_platesmania"),
datasets.BuilderConfig(name="Kazakhstan_youtube"),
datasets.BuilderConfig(name="Lithuania_platesmania"),
datasets.BuilderConfig(name="Serbia_platesmania"),
datasets.BuilderConfig(name="Serbia_youtube"),
datasets.BuilderConfig(name="UAE_platesmania"),
datasets.BuilderConfig(name="UAE_youtube")
]
DEFAULT_CONFIG_NAME = "Brazil"
def _info(self):
features = datasets.Features({
'bbox_id': datasets.Value('uint32'),
'bbox': datasets.Value('string'),
'image': datasets.Image(),
'labeled_image': datasets.Image(),
'license_plate.id': datasets.Value('string'),
'license_plate.visibility': datasets.Value('string'),
'license_plate.rows_count': datasets.Value('uint8'),
'license_plate.number': datasets.Value('string'),
'license_plate.serial': datasets.Value('string'),
'license_plate.country': datasets.Value('string'),
'license_plate.mask': datasets.Value('string')
})
return datasets.DatasetInfo(
description=_DESCRIPTION,
features=features,
supervised_keys=None,
homepage=_HOMEPAGE,
citation=_CITATION,
)
def _split_generators(self, dl_manager):
data = dl_manager.download(f"{_DATA}{self.config.name}.tar.gz")
data = dl_manager.iter_archive(data)
annotations = dl_manager.download(f'{_DATA}{self.config.name}.csv')
return [
datasets.SplitGenerator(name=datasets.Split.TRAIN,
gen_kwargs={
"data": data,
'annotations': annotations
}),
]
def _generate_examples(self, data, annotations):
annotations_df = pd.read_csv(annotations, sep=',', index_col=0)
images = {}
for idx, (file_path, file) in enumerate(data):
file_name = file_path.split('/')[-1]
images[file_name] = (file_path, file.read())
annotations_df.drop(
columns=['license_plate.region', 'license_plate.color'],
inplace=True,
errors='ignore')
annotations_df.fillna(0, inplace=True)
annotations_df.sort_values(by='file_name', inplace=True)
for row in annotations_df.itertuples(index=True):
image = images[row[1]]
name, ext = row[1].split('.')
labeled_image = images[f'{name}_labeled.{ext}']
yield idx, {
'bbox_id': row[0],
'bbox': row[2],
"image": {
"path": image[0],
"bytes": image[1]
},
"labeled_image": {
"path": labeled_image[0],
"bytes": labeled_image[1]
},
'license_plate.id': row[3],
'license_plate.visibility': row[4],
'license_plate.rows_count': row[5],
'license_plate.number': row[6],
'license_plate.serial': row[7],
'license_plate.country': row[8],
'license_plate.mask': row[9]
}