README.md

metadata

dataset_info:
  features:
    - name: image
      dtype: image
    - name: label
      dtype:
        class_label:
          names:
            '0': body_swing
            '1': crawling
            '2': jumping
            '3': looking_left
            '4': looking_right
            '5': looking_up_or_down
            '6': moving_down
            '7': moving_up
            '8': sitting
            '9': standing
            '10': turning_around
            '11': turning_left
            '12': turning_right
            '13': walking
  splits:
    - name: train
      num_bytes: 5145133631.037
      num_examples: 31703
    - name: validation
      num_bytes: 1236821917.354
      num_examples: 7918
  download_size: 5783153535
  dataset_size: 6381955548.391
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
task_categories:
  - image-classification
size_categories:
  - 10K<n<100K

Kortowo Dynamics Image Dataset

This is the image dataset for the Kortowo Dynamics project.

It contains frames derived from the original video dataset.

Purpose

The initial purpose of this dataset has been to train a video classification model using the strategy of analysing individual frames.

Contents

The dataset contains video frames of the Boston Dynamics' Spot robot performing different actions.

Available classes:

• body_swing
• crawling
• jumping
• looking_left
• looking_right
• looking_up_or_down
• moving_down
• moving_up
• sitting
• standing
• turning_around
• turning_left
• turning_right
• walking