Main Menu

Home
Organizers
Important Dates
Results
Contest Description
Evaluation Criteria
Download
Participation

Download datasets

The details of the datasets are presented in the following publication:

Atsushi Shimada, Kazuaki Kondo, Daisuke Deguchi, Géraldine Morin, Helman Stern, "Kitchen Scene Context Based Gesture Recognition: A Contest in ICPR2012," Advances in Depth Image Analysis and Applications, Lecture Notes in Computer Science Volume 7854, pp.168-185, 2013.
If you intend to use the datasets, please cite the above paper within your publication.

Overview

There are five candidate cooking menus. Each menu is cooked by seven different actors (five actors in training datasets, and two actors in evaluation datasets); that is, five cooking scenes are available for each menu. Detailed descriptions of the datasets are available in contest description page.

In this contest, we will provide training datasets and evaluation datasets. Currently, we can provide five datasets for training, and two datasets for evaluation.

After downloading datasets, please check a label description file. This file provides you an information about labels' meaning used in the datasets. Please use these labels to train your programs.

When submitting the results, it is necessary to follow the submission guideline described in the participation page.

If you have any questions, please contact to us via mailing list.

Specification of datasets

Each cooking video was divided into image sequence files (color image sequences and depth image sequences), and each file name represents its captured time in usec format.

A label no in the label file corresponds to an image sequence file in ascending order.

Each depth image consists of 8-bit gray scale information (256 levels). The depth unit of each level is 8mm. Therefore, the depth was available up to 2048mm (8mm x 256level).

Training Datasets

Data1, data2, ..., data5 are the datasets corresponding to individual actors, and each data contains five types of cooking menus.

Data 1
Video: data1.tar.gz [3.7GB], Label: label1.tar.gz [105KB]
Data 2
Video: data2.tar.gz [2.8GB], Label: label2.tar.gz [79KB]
Data 3
Video: data3.tar.gz [1.8GB], Label: label3.tar.gz [49KB]
Data 4
Video: data4.tar.gz [3.3GB], Label: label4.tar.gz [94KB]
Data 5
Video: data5.tar.gz [2.9GB], Label: label5.tar.gz [79KB]


Evaluation Datasets

Following data10 and data11 are the evaluation datasets, and they are corresponding to individual actors, and each data contains five types of cooking menus.

To submit the results, contestants have to follow the instructions described in the participation page.

Data 10
Video: data10.tar.gz [2.2GB], Label: label10.tar.gz [63KB]
Data 11
Video: data11.tar.gz [2.4GB], Label: label11.tar.gz [68KB]