Hugging Face's logo

Datasets:
khaclinh
/
testdata

Tasks:
Object Detection
Sub-tasks:
face-detection
Languages:
English
Size:
1K<n<10K
License:
Dataset card Files Community
2
testdata / README.md
khaclinh's picture
khaclinh
Update README.md
e9b5e2c
preview code
|
raw
history blame
8.79 kB
metadata
annotations_creators:
  - expert-generated
language_creators:
  - found
language:
  - en
license:
  - cc-by-nc-nd-4.0
multilinguality:
  - monolingual
size_categories:
  - 1K<n<10K
source_datasets:
  - extended
task_categories:
  - object-detection
task_ids:
  - face-detection
  - license-plate-detection
pretty_name: PP4AV

Dataset Card for PP4AV

Table of Contents

Dataset Description

Dataset Summary

PP4AV is the first public dataset with faces and license plates annotated with driving scenarios. P4AV provides 3,447 annotated driving images for both faces and license plates. For normal camera data, dataset sampled images from the existing videos in which cameras were mounted in moving vehicles, running around the European cities. The images in PP4AV were sampled from 6 European cities at various times of day, including nighttime. This dataset use the fisheye images from the WoodScape dataset to select 244 images from the front, rear, left, and right cameras for fisheye camera data. PP4AV dataset can be used as a benchmark suite (evaluating dataset) for data anonymization models in autonomous driving.

Supported Tasks and Leaderboards

  • face-detection: The dataset can be used to train a model for Face Detection. More information on evaluating the model's performance can be found here.

Languages

English

Dataset Structure

Data Instances

A data point comprises an image and its face and license plate annotations.

{
  'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=1024x755 at 0x19FA12186D8>, 'objects': {
    'bbox': [
      [0 0.230078 0.317081 0.239062 0.331367],
      [1 0.5017185 0.0306425 0.5185935 0.0410975],
      [1 0.695078 0.0710145 0.7109375 0.0863355],
      [1 0.4089065 0.31646 0.414375 0.32764],
      [0 0.1843745 0.403416 0.201093 0.414182],
      [0 0.7132 0.3393474 0.717922 0.3514285]
    ]
  }
}

Data Fields

  • image: A PIL.Image.Image object containing the image. Note that when accessing the image column: dataset[0]["image"] the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the "image" column, i.e. dataset[0]["image"] should always be preferred over dataset["image"][0]
  • objects: a dictionary of face and license plate bounding boxes present on the image
    • bbox: the bounding box of each face and license plate (in the yolo format). Basically, each row in annotation .txt file for each image .png file consists of data in format: <object-class> <x_center> <y_center> <width> <height>:
      • object-class: integer number of object from 0 to 1, where 0 indicate face object, and 1 indicate licese plate object
      • x_center: normalized x-axis coordinate of the center of the bounding box.
        x_center = <absolute_x_center> / <image_width>
      • y_center: normalized y-axis coordinate of the center of the bounding box.
        y_center = <absolute_y_center> / <image_height>
      • width: normalized width of the bounding box.
        width = <absolute_width> / <image_width>
      • height: normalized wheightdth of the bounding box.
        height = <absolute_height> / <image_height>
      • Example lines in YOLO v1.1 format .txt' annotation file: 1 0.716797 0.395833 0.216406 0.147222
        0 0.687109 0.379167 0.255469 0.158333
        1 0.420312 0.395833 0.140625 0.166667 `

Dataset Creation

Curation Rationale

The curators state that the current face detection datasets typically contain a few thousand faces, with limited variations in pose, scale, facial expression, occlusion, and background clutters, making it difficult to assess for real world performance. They argue that the limitations of datasets have partially contributed to the failure of some algorithms in coping with heavy occlusion, small scale, and atypical pose.

Source Data

Initial Data Collection and Normalization

The objective of PP4AV is to build a benchmark dataset that can be used to evaluate face and license plate detection models for autonomous driving. For normal camera data, we sampled images from the existing videos in which cameras were mounted in moving vehicles, running around the European cities. We focus on sampling data in urban areas rather than highways in order to provide sufficient samples of license plates and pedestrians. The images in PP4AV were sampled from 6 European cities at various times of day, including nighttime. We use the fisheye images from the WoodScape dataset to select 244 images from the front, rear, left, and right cameras for fisheye camera data. In total, 3,447 images were selected and annotated in PP4AV.

Who are the source language producers?

The images are selected from publicly available WIDER dataset.

Annotations

Annotation process

Annotators annotate facial and license plate objects in images. For facial objects, bounding boxes are defined by all detectable human faces from the forehead to the chin to the ears. Faces were labelled with diverse sizes, skin tones, and faces partially obscured by a transparent material, such as a car windshield. For license plate objects, bounding boxes consists of all recognizable license plates with high variability, such as different sizes, countries, vehicle types (motorcycle, automobile, bus, truck), and occlusions by other vehicles. License plates were annotated for vehicles involved in moving traffic. To ensure the quality of annotation, there are two-step process for annotation. In the first phase, two teams of annotators will independently annotate identical image sets. After their annotation output is complete, a merging method based on the IoU scores between the two bounding boxes of the two annotations will be applied. Pairs of annotations with IoU scores above a threshold will be merged and saved as a single annotation. Annotated pairs with IoU scores below a threshold will be considered conflicting. In the second phase, two teams of reviewers will inspect the conflicting pairs of annotations for revision before a second merging method similar to the first is applied. The results of these two phases will be combined to form the final annotation. All work is conducted on the CVAT tool https://github.com/openvinotoolkit/cvat.

Who are the annotators?

Vantix Data Science team

Personal and Sensitive Information

[More Information Needed]

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

Linh Trinh

Licensing Information

Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).

Citation Information 

@article{PP4AV2022,
  title = {PP4AV: A benchmarking Dataset for Privacy-preserving Autonomous Driving},
  author = {Linh Trinh, Phuong Pham, Hoang Trinh, Nguyen Bach, Dung Nguyen, Giang Nguyen, Huy Nguyen},
  booktitle = {IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year = {2023}
}

Contributions

Thanks to @khaclinh for adding this dataset.