shapestacks

ShapeStacks

ShapeStacks teaser image

ShapeStacks Dataset

document icon ShapeStacks-Manual.md

MuJoCo world definitions (39 MB):
download icon shapestacks-mjcf.tar.gz · shapestacks-mjcf.md5

Meta information (156 KB):
download icon shapestacks-meta.tar.gz · shapestacks-meta.md5

RGB images (33 GB):
download icon shapestacks-rgb.tar.gz · shapestacks-rgb.md5

Violation segmentation maps (875 MB):
download icon shapestacks-vseg.tar.gz · shapestacks-vseg.md5

Instance segmentation maps (352 MB):
download icon shapestacks-iseg.tar.gz · shapestacks-iseg.md5

Depth maps (1.1 GB):
download icon shapestacks-depth.tar.gz · shapestacks-depth.md5

The dataset can also be downloaded from the command line using wget:

wget http://shapestacks-file.robots.ox.ac.uk/static/download/v1/ShapeStacks-Manual.md  # ShapeStacks-Manual.md
wget http://shapestacks-file.robots.ox.ac.uk/static/download/v1/shapestacks-mjcf.tar.gz  # shapestacks-mjcf.tar.gz
wget http://shapestacks-file.robots.ox.ac.uk/static/download/v1/shapestacks-meta.tar.gz  # shapestacks-meta.tar.gz
wget http://shapestacks-file.robots.ox.ac.uk/static/download/v1/shapestacks-rgb.tar.gz  # shapestacks-rgb.tar.gz
wget http://shapestacks-file.robots.ox.ac.uk/static/download/v1/shapestacks-vseg.tar.gz  # shapestacks-iseg.tar.gz
wget http://shapestacks-file.robots.ox.ac.uk/static/download/v1/shapestacks-iseg.tar.gz  # shapestacks-iseg.tar.gz
wget http://shapestacks-file.robots.ox.ac.uk/static/download/v1/shapestacks-depth.tar.gz  # shapestacks-depth.tar.gz

Source Code

code icon Check out the source code on GitHub.

We provide a dataset handler for ShapeStacks using Tensorflow dataset API. It also contains example scripts for training and evaluation of the stablity prediction models.

Pre-trained stability prediction models based on InceptionV4 (1.8 GB):
download icon shapestacks-incpv4.tar.gz · shapestacks-incpv4.md5

Paper @ ECCV 2018

ShapeStacks paper teaser

document icon [paper] image icon [poster] video icon [video]

Physical intuition is pivotal for intelligent agents to perform complex tasks. In this paper we investigate the passive acquisition of an intuitive understanding of physical principles as well as the active utilisation of this intuition in the context of generalised object stacking. To this end, we provide ShapeStacks: a simulation-based dataset featuring 20,000 stack configurations composed of a variety of elementary geometric primitives richly annotated regarding semantics and structural stability.

Contact

team labs labs
Oliver Groth, Fabian Fuchs, Ingmar Posner, Andrea Vedaldi Applied AI Lab, Visual Geometry Group Research funded by ERC 677195-IDIU, AIMS-CDT