Datasets:

Modalities:
Image
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
License:
Wake-Vision / README.md
davanstrien's picture
davanstrien HF staff
add minimal dataset card with link to paper
e2dc14f verified
|
raw
history blame
8.41 kB
metadata
license: cc-by-4.0
dataset_info:
  features:
    - name: age_unknown
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: body_part
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: bright
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: dark
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: depiction
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: far
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: filename
      dtype: string
    - name: gender_unknown
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: image
      dtype: image
    - name: medium_distance
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: middle_age
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: near
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: non-person_depiction
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: non-person_non-depiction
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: normal_lighting
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: older
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: person
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: person_depiction
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: predominantly_female
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: predominantly_male
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
    - name: young
      dtype:
        class_label:
          names:
            '0': 'No'
            '1': 'Yes'
  splits:
    - name: test
      num_bytes: 15119280526
      num_examples: 53304
    - name: validation
      num_bytes: 5013154770.625
      num_examples: 17627
  download_size: 20127967346
  dataset_size: 20132435296.625
configs:
  - config_name: default
    data_files:
      - split: train_quality
        path: data/train_quality*
      - split: test
        path: data/test-*
      - split: validation
        path: data/validation-*
task_categories:
  - image-classification
pretty_name: Wake Vision
size_categories:
  - 1M<n<10M

Dataset Card for Wake Vision

Dataset Details

Dataset Description

Paper abstract:

Abstract. Machine learning applications on extremely low-power de- vices, commonly referred to as tiny machine learning (TinyML), promises a smarter and more connected world. However, the advancement of cur- rent TinyML research is hindered by the limited size and quality of per- tinent datasets. To address this challenge, we introduce Wake Vision, a large-scale, diverse dataset tailored for person detection—the canonical task for TinyML visual sensing. Wake Vision comprises over 6 million images, which is a hundredfold increase compared to the previous stan- dard, and has undergone thorough quality filtering. Using Wake Vision for training results in a 2.41% increase in accuracy compared to the estab- lished benchmark. Alongside the dataset, we provide a collection of five detailed benchmark sets that assess model performance on specific seg- ments of the test data, such as varying lighting conditions, distances from the camera, and demographic characteristics of subjects. These novel fine-grained benchmarks facilitate the evaluation of model quality in chal- lenging real-world scenarios that are often ignored when focusing solely on overall accuracy. Through an evaluation of a MobileNetV2 TinyML model on the benchmarks, we show that the input resolution plays a more crucial role than the model width in detecting distant subjects and that the impact of quantization on model robustness is minimal, thanks to the dataset quality. These findings underscore the importance of a de- tailed evaluation to identify essential factors for model development. The dataset, benchmark suite, code, and models are publicly available under the CC-BY 4.0 license, enabling their use for commercial use cases

  • Curated by: [More Information Needed]
  • Funded by [optional]: [More Information Needed]
  • Shared by [optional]: [More Information Needed]
  • Language(s) (NLP): [More Information Needed]
  • License: [More Information Needed]

Dataset Sources [optional]

  • Repository: [More Information Needed]
  • Paper [optional]: [More Information Needed]
  • Demo [optional]: [More Information Needed]

Uses

Direct Use

[More Information Needed]

Out-of-Scope Use

[More Information Needed]

Dataset Structure

[More Information Needed]

Dataset Creation

Curation Rationale

[More Information Needed]

Source Data

Data Collection and Processing

[More Information Needed]

Who are the source data producers?

[More Information Needed]

Annotations [optional]

Annotation process

[More Information Needed]

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

[More Information Needed]

Bias, Risks, and Limitations

[More Information Needed]

Recommendations

Citation

BibTeX:

@misc{banbury2024wake,
      title={Wake Vision: A Large-scale, Diverse Dataset and Benchmark Suite for TinyML Person Detection}, 
      author={Colby Banbury and Emil Njor and Matthew Stewart and Pete Warden and Manjunath Kudlur and Nat Jeffries and Xenofon Fafoutis and Vijay Janapa Reddi},
      year={2024},
      eprint={2405.00892},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

[More Information Needed]

Dataset Card Contact

[More Information Needed]