ETH ZURICH | | CALVIN group

Synchronic Activities Stickmen V 1.0

Annotated data and evaluation routines for 2D human pose estimation



Marcin Eichner, Vittorio Ferrari



Annotated example Annotated example Annotated example Annotated example


Overview

Dataset sticks distribution
Dataset sticks distribution

Welcome to this release of the Synchronic Activities Stickmen dataset! This dataset focuses on scenarios where multiple persons perform some activity synchronously. This happens often in sports (gymnastics, diving) or when a group of people exercises following a leader (aerobic, cheerleading, dancing, martial arts). We release here a dataset of such synchronic activity photos of dancing, aerobic and cheer-leading, complete with annotations of the six upper body parts of all persons performing the activity. The dataset has 357 images sampled from synchronic activities videos downloaded from Youtube.

In each image we have annotated the upper-body of every approximately upright and frontal person, for a total of 1128 persons (on average 3 persons per image). A body part is annotated by a line segment. The parts annotated are head, torso, upper and lower arms.

Results on this dataset have been first published in [1]. Please cite it if you use this dataset.

In addition, the package includes official Matlab routines to evaluate the performance of your pose estimation system on this dataset and compare to our results from [1].

On the right, the scatter plot inspired by [2] depicts pose variability over this dataset. Stickmen are centered on the neck and scale normalized. Hence the plot captures only pose variability and does not show scale and location variability.

Clarification of the PCP evaluation criterion

The matlab code to evaluate PCP provided with this dataset represents the official evaluation protocol for the following datasets: Buffy Stickmen, ETHZ PASCAL Stickmen, We Are Family Stickmen, Synchronic Activities Stickmen. In our PCP implementation, a body part produced by an algorithm is considered correctly localized if its endpoints are closer to their ground-truth locations than a threshold (on average over the two endpoints). Using it ensures results comparable to the vast majority of results previously reported on these datasets... Read more

Downloads

FilenameDescriptionSize
SynchronicActivities_Stickmen_v1.0.tgz annotations for the included frames and matlab code to read and display the annotations and evaluate pose estimation performance. 29 MB
README.txt description of contents. 14 kB
PCP_pami12_SynchronicActivities_Results.png plot showing pose estimation performance of [1] on this dataset 32 kB
PCP_pami12_SynchronicActivities_Results.fig Matlab figure plot. You can overlay your performance curve on this plot in order to compare to our results from [1] 18 kB

Related Publications

[1] Eichner, M. and Ferrari, V.
Human Pose Co-Estimation and Applications
IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), (to appear)
Document: PDF

[2] D.Tran, D.Forsyth
Improved Human Parsing with a Full Relational Model
Proceedings of European Conference on Computer Vision (ECCV) 2010

Acknowledgements

This work is funded the Swiss National Science Foundation SNSF

Please report problems with this page to
Marcin Eichner
Last updated on Tuesday, 02nd February, 2016