image
image
128
128
label
class label
2 classes
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting
0biting

Dataset Card for Nail Biting Classification

Dataset Summary

A binary image dataset for classifying nailbiting. Images are cropped to only show the mouth area. Should contain edge cases such as drinking water, talking on the phone, scratching chin etc.. all in "no biting" category

Dataset Structure

Data Instances

  • 7147 Images
  • 14879790 bytes total
  • 12332617 bytes download

Data Fields

128 x 64 (w x h, pixels) Black and white

Labels

  • '0': biting
  • '1': no_biting

Data Splits

  • train: 6629 (11965737 bytes)
  • test: 1471 (2914053 bytes)

Dataset Creation

Curation Rationale

I wanted to create a notification system to help me stop biting my nails. It needed to contain lots of possible no-biting scenarios. eg talking on the phone

Source Data

Initial Data Collection and Normalization

The data was scraped from stock images sites and photos of myself were taken with my webcam. MTCNN (https://github.com/ipazc/mtcnn) was then used to crop the images down to only the show the mouth area The images were then converted to a black & white colour scheme.

Annotations

Annotation process

During the scraping process images were labelled with a description, which I then manually sanity checked. I labelled the ones of me manually.

Who are the annotators?

Alec Sharp

Considerations for Using the Data

Discussion of Biases & Limitations

Tried to make the dataset diverse in terms of age and skin tone. Although, this dataset contains a large number of images of one subject (me) so is biased towards lower quality webcam pictures of a white male with a short beard.

Dataset Curators

Alec Sharp

Licensing Information

MIT

Contributions

Thanks to @alecsharpie for adding this dataset.

Downloads last month
2