===========================================================================
vNicoWorld-LL (virtual NICO World for Lifelong Learning dataset)
===========================================================================


Introduction
------------
- An object recognition dataset particularly created for lifelong
  (continual) learning. It was recorded in a nearly real-world virtual
  environment in blender and consists of 95 objects belonging to 10
  categories. The categories are:

  -- ball
  -- book
  -- bottle
  -- cup
  -- doughnut
  -- glasses
  -- pen
  -- pocket watch
  -- present
  -- vase

- A virtual version of the humanoid robot NICO manipulated
  these objects in front of 20 different backgrounds. To simulate
  different environments, real-world and artificially created background
  images, grouped into four different complexity levels (B1, B2, B3, B4),
  are utilized.

- Hence, the dataset consists of 95*20=1900 videos and each video of 240
  frames.


General Information
-------------------
- The folder "data" contains the dataset. The general folder structure is:

    |-- v_NICO_World_LL
    |   |-- s*
    |   |   |-- o*
    |   |   |   |-- C_**_**_****.png

- The 20 different "s*" folders (s0-s19) represent the different
  backgrounds/sessions.

- The 95 different "o*" folders (o0-o99) represent the different objects.

- Each file represents a single frame in a video. The frame order is given
  by the last four numbers of the filename (0000-0240).

- The "background_complexities.json" file contains the information of which
  background/session belongs to which background complexity
  (B1, B2, B3, B4).

- Most objects are modified by the author to fit into the hand of the
  robot and to create a higher diversity.


Authors
-------
- Name: Aleksej Logacjov
- E-Mail 1: a.logacjov@gmail.com


===========================================================================
Date: 2022-04-11