HaDa Dataset

The HaDa (Hands dataset) contains depth and color images of shopper-shelf interaction from top-view configuration. In particular, the purpose of this dataset is to classify the images in these categories:

  • Positive (image contains a hand holding something)


  • Negative (image contains a empty hand)


  • Neutral (none of the other)


  • Refill (shelf refilling operation)


An interaction is described by the sequence of two images.

It contains a total of 13854 images (depth + color) with a dimension of 70 x 70 pixels.The ground truth was manually labelled by human annotators. The following figure shows an example of a dataset instances that includes the four categories described above.

Note that the dataset is currently under review and available only for research purposes. Delay in responses may occurs.

  • Send an email to: vrai@dii.univpm.it (Note: you should send the email from an email address that is linked to your research institution/university)
  • You will be sent a link for the download.