Primitives Dataset

A small dataset consisting of short video clips, each containing one of six shapes (cone, cube, cylinder, monkey, sphere, torus) on a black background.

The dataset was used during my experiments with Hierarchical Temporal Memory involving video classification ​(Wielgosz and Pietroń 2017)​.

The clips were generated using Blender. Available clip lengths are 4 frames (10k clips), 32 frames (1k clips), and 150 frames (1k clips). Additionally, the 32-frames-long clips are available in five resolutions. Accompanying video clips is an image dataset (10k images).

The dataset is available under Creative Commons Attribution 4.0 International License.

Example frames from primitives dataset
  1. Wielgosz, Maciej, and Marcin Pietroń. 2017. “Using Spatial Pooler of Hierarchical Temporal Memory to Classify Noisy Videos with Predefined Complexity.” Neurocomputing, May, 84–97.
Maciej Wielgosz's Avatar
Written by

Maciej Wielgosz