SignSeq-dataset / README.md
JacobLinCool's picture
Upload dataset
bef39a3 verified
metadata
dataset_info:
  features:
    - name: created_at
      dtype: string
    - name: sequence
      sequence:
        sequence: float64
    - name: width
      dtype: float64
    - name: height
      dtype: float64
    - name: preview_image
      dtype: image
    - name: num_points
      dtype: int64
    - name: duration
      dtype: float64
    - name: avg_force
      dtype: float64
    - name: min_force
      dtype: float64
    - name: max_force
      dtype: float64
    - name: avg_altitude
      dtype: float64
    - name: min_altitude
      dtype: float64
    - name: max_altitude
      dtype: float64
    - name: avg_azimuth_sin
      dtype: float64
    - name: min_azimuth_sin
      dtype: float64
    - name: max_azimuth_sin
      dtype: float64
    - name: avg_azimuth_cos
      dtype: float64
    - name: min_azimuth_cos
      dtype: float64
    - name: max_azimuth_cos
      dtype: float64
    - name: signer
      dtype: string
  splits:
    - name: train
      num_bytes: 10848294
      num_examples: 137
  download_size: 8296862
  dataset_size: 10848294
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
extra_gated_description: >-
  Due to the consent forms we use, we cannot share the dataset with others. But
  we may be able to help other researchers seek permission from the signatories.
extra_gated_fields:
  Country: country
  Institution: text
  I want to use this dataset for: text

SignSeq Dataset

The SignSeq dataset contains time-series handwriting signature traces collected at 240 Hz by Apple Pencil on iPad. Each example records the full stroke sequence along with per-stroke metadata and a rendered preview image. Intended use cases include signature generation, style analysis, and handwriting modelling.

Features

Each record contains the following fields:

Field Type Description
created_at string ISO-8601 timestamp of recording start.
sequence sequence<sequence<float>> A list of points: each [t, x_rel, y_rel, force, altitude, azimuth_sin, azimuth_cos].
width float64 Original signature width. Used to de-normalize x_rel back to pixel coordinates.
height float64 Original signature height. Used to de-normalize y_rel back to pixel coordinates.
preview_image image PNG rendering of the full signature trace.
num_points int64 Total number of captured points in sequence.
duration float64 Recording duration in seconds.
avg_force float64 Mean pen pressure over the signature.
min_force float64 Minimum pen pressure.
max_force float64 Maximum pen pressure.
avg_altitude float64 Mean pen altitude angle (radians).
min_altitude float64 Minimum pen altitude angle (radians).
max_altitude float64 Maximum pen altitude angle (radians).
avg_azimuth_sin float64 Mean sine of pen azimuth angle (radians).
min_azimuth_sin float64 Minimum sine of pen azimuth angle (radians).
max_azimuth_sin float64 Maximum sine of pen azimuth angle (radians).
avg_azimuth_cos float64 Mean cosine of pen azimuth angle (radians).
min_azimuth_cos float64 Minimum cosine of pen azimuth angle (radians).
max_azimuth_cos float64 Maximum cosine of pen azimuth angle (radians).
signer string Anonymized signer identifier.

Notes on sequence

  • Element breakdown
    1. t : Relative timestamp in seconds (starts at 0.0 for the first point).
    2. x_rel, y_rel : Relative coordinates in [0, 1], to be multiplied by width/height to recover pixel positions.
    3. force : Pen pressure (unitless).
    4. altitude : Pen altitude angle in radians.
    5. azimuth_sin, azimuth_cos : Sine and cosine of pen azimuth angle in radians.