HandNet 数据集

HandNet

Overview

The HandNet dataset contains depth images of 10 participants' hands non-rigidly deforming infront of a RealSense RGB-D camera. The annotations are generated by a magnetic annotation technique as described in our paper referenced below. 6D pose is available for the center of the hand as well as the five fingertips (i.e. position and orientation of each).

License

You may use the files of this database and related code for academic, non-commerical purposes. For any other use please contact the authors directly at: twerd {at} cs.technion.ac.il

Download

The files can be downloaded from the following links. Because of the large size of the training data it is suggested to first download the testing and/or validation data as well as the code on Github to determine whether the data is suitable for your purposes.

Train data (202198 images)
Test data (10000 images)
Validation data (2773 images)
Train_DerotGT (202198 images) 
Validation_DerotGT (2773 images)
Test_Perturbed (10000 images)
Validation_Perturbed (2773 images) 
Code

Download (12.5 GB)
Download (627 MB)
Download (174 MB)
Download (10 GB)
Download (142 MB)
Download (521 MB)
Download (144 MB)
Github

References

The creation of the Handnet database is partially described in the following paper. Please cite this work using the bibtex link below if you use the database in your own work.

Rule of Thumb: Deep Derotation for Improved Fingertip Detection,
Aaron Wetzler, Ron Slossberg and Ron Kimmel, BMVC 2015 

[Paper]   [Poster]   [Supplementary material]   [Bibtex]

HandNet benchmark

HandNet can be used to benchmark hand pose methods and various machine learning methods. If you would like to submit your results on the Test dataset (the ground truth is available in the dataset) then please send an email to twerd {at} cs.technion.ac.il. The code for benchmarking can be found on github.

Benchmark results

Method	Results mP | mAP	Authors	Description	Evaluation date
RDT-Baseline RDT-DeROT CNN-Baseline CNN-DeROT	0.51 | 0.79 0.63 | 0.88 0.44 | 0.73 0.61 | 0.88	A Wetzler, R Slossberg, R Kimmel.  GIP lab, Technion	Based on Rule of Thumb: Deep derotation  for improved fingertip detection	05/09/2015

Affiliation

The database was created as part of research carried out at the GIP lab in the Technion Computer Science faculty. It was partially supported by the European Community’s FP7-ERC program, grant agreement no. 267414

from: http://www.cs.technion.ac.il/~twerd/HandNet/