MANIAC Dataset

We provide a manipulation action dataset with 8 different manipulation actions, each of which consists of 15 different versions performed by 5 different human actors. There are in total 30 different objects manipulated in all demonstrations. All manipulations were recorded with the Microsoft Kinect sensor.

We extend our data set with 20 long and complex chained manipulation sequences (e.g. “making a sandwich”) which consist of in total 103 different versions of these 8 manipulation tasks performed in different orders with novel objects under different circumstances.

Please cite the following paper in the case of using this dataset in your publications (bibtex, pdf): Eren Erdal Aksoy, Tamosiunaite Minija, and Wörgötter Florentin (2014). “Model-Free Incremental Learning of the Semantics of Manipulation Actions”. Robotics and Autonomous Systems (In press).

Project Contact: Dr. Eren Erdal Aksoy

Download

A readme file on how to use the dataset can be found here.

Action Movie PCD file SECs Labels
Pushing Actions (mpg, 9 MB) (pcd, 8 GB) (xml, 1 MB) (dat, 12 MB)
Putting Actions (mpg, 9 MB) (pcd, 8 GB) (xml, 1 MB) (dat, 12 MB)
Hiding Actions (mpg, 11 MB) (pcd, 11 GB) (xml, 2 MB) (dat, 16 MB)
Stirring Actions (mpg, 18 MB) (pcd, 22 GB) (xml, 4 MB) (dat, 39 MB)
Cutting Actions (mpg, 17 MB) (pcd, 19 GB) (xml, 4 MB) (dat, 38 MB)
Chopping Actions (mpg, 15 MB) (pcd, 18 GB) (xml, 4 MB) (dat, 35 MB)
Taking Actions (mpg, 69 GB) (pcd, 11 MB) (xml, 2 MB) (dat, 20 MB)
Uncovering Actions (mpg, 12 MB) (pcd, 12 GB) (xml, 2 MB) (dat, 20 MB)
Chained Actions Movie PCD file SECs Labels
Version 1 (mpg, 2 MB) (pcd, 2 GB) (xml, 1 MB) (dat, 4 MB)
Version 2 (mpg, 2 MB) (pcd, 2 GB) (xml, 1 MB) (dat, 4 MB)
Version 3 (mpg, 2 MB) (pcd, 2 GB) (xml, 1 MB) (dat, 4 MB)
Version 4 (mpg, 4 MB) (pcd, 8 GB) (xml, 1 MB) (dat, 17 MB)
Version 5 (mpg, 1 MB) (pcd, 1 GB) (xml, 2 MB) (dat, 1 MB)
Version 6 (mpg, 1 MB) (pcd, 2 GB) (xml, 1 MB) (dat, 2 MB)
Version 7 (mpg, 1 MB) (pcd, 2 GB) (xml, 1 MB) (dat, 4 MB)
Version 8 (mpg, 3 MB) (pcd, 4 GB) (xml, 1 MB) (dat, 10 MB)
Version 9 (mpg, 2 MB) (pcd, 3 GB) (xml, 1 MB) (dat, 4 MB)
Version 10 (mpg, 1 MB) (pcd, 3 GB) (xml, 1 MB) (dat, 6 MB)
Version 11 (mpg, 2 MB) (pcd, 3 GB) (xml, 1 MB) (dat, 6 MB)
Version 12 (mpg, 2 MB) (pcd, 3 GB) (xml, 1 MB) (dat, 5 MB)
Version 13 (mpg, 2 MB) (pcd, 3 GB) (xml, 1 MB) (dat, 7 MB)
Version 14 (mpg, 2 MB) (pcd, 4 GB) (xml, 1 MB) (dat, 8 MB)
Version 15 (mpg, 2 MB) (pcd, 4 GB) (xml, 1 MB) (dat, 8 MB)
Version 16 (mpg, 3 MB) (pcd, 5 GB) (xml, 1 MB) (dat, 9 MB)
Version 17 (mpg, 3 MB) (pcd, 5 GB) (xml, 1 MB) (dat, 11 MB)
Version 18 (mpg, 5 MB) (pcd, 11 GB) (xml, 2 MB) (dat, 26 MB)
Version 19 (mpg, 5 MB) (pcd, 9 GB) (xml, 2 MB) (dat, 23 MB)
Version 20 (mpg, 1 MB) (pcd, 2 GB) (xml, 1 MB) (dat, 4 MB)

Readme

The dataset has two parts:

8 atomic manipulation actions, each has 15 versions. 20 chained manipulations.

For each recorded manipulation we provide the raw orginazed pcl data (xyzrgb). Attached movies show how each manipulation looks like. We also provide segment labels that were extraced from RGBD pcl data using the method in Abramov 2012 (see below). The label set for each frame was saved in a matrix format in the respective .dat file. File names between pcl and segment labels are compatible!

In order to get the respective labeled point cloud, one has to simply assign labels to points in the cloud. We additionally provide the extracted Semantic Event Chain (SEC) data. In each SEC folder, SEC data are provided as a GraphML file with respective key frame images. These key frame images are just for the visualization. Upon the request, we can also provide an xml parser that converts SEC GraphML files into Matlab matrices. Inside the SEC folder, the SEC data was also provided in a matrix format as .dat file.

Please feel free to contact Eren Erdal Aksoy (aksoyeren@gmail.com) if there is anything missing!

Please cite the following paper in the case of using this dataset in your publications: Eren Erdal Aksoy, Tamosiunaite Minija, and Wörgötter Florentin (2014). “Model-Free Incremental Learning of the Semantics of Manipulation Actions”. Robotics and Autonomous Systems (In press).

Segmentation paper: A. Abramov, K. Pauwels, J. Papon, F. Wörgötter, B. Dellen, Depth supported real-time video segmentation with the kinect, in: Applications of Computer Vision (WACV), 2012 IEEE Workshop on, 2012, pp. 457-464.

Computational Neuroscience Group