Home
Uni-Logo
 

Datasets

Binaries/Code Datasets Open Source Software

Freiburg-Berkeley Motion Segmentation Dataset
Video Segmentation Benchmark
Image Sequences
TEM Dataset
TILDA Textile Texture Database
Training data for Exemplar CNN
Generated Matching Dataset
Training data for chair generation
Stereo Ego-motion Dataset
Optical Flow Datasets: "Flying Chairs", "ChairsSDHom"
Scene Flow Datasets
Human Part Segmentation Datasets  
Rendered Handpose Dataset
Pedestrian Zone Scene
FreiHAND Dataset
HanCo Dataset
Human Pose RGBD Datasets
OVAD: Open-Vocabulary Attribute Detection Dataset
ADE-OoD: a benchmark for dense Out-of-Distribution detection on natural images.


HanCo - A Collection of Hand Images for Unsupervised Representation Learning

In our recent publication we presented HanCo a dataset that allowed us to surpass ImageNet based pretraining for the task of Hand Shape Estimation from a single RGB image. HanCo is a structured collection of hand images which makes it an ideal testbench for unsupervised representation learning. The collection was aquired by recording short video clips with a calibrated and time sychronized multi-view camera capture setup with a frame rate of 5Hz. This dataset allows to develop unsupervised learning approaches that exploit time and view-based consistency contraints as well as leveraging exact multi-view geometry. Our dataset provides:
- RGB images (224x224 pixels)
- Foreground segmentation (Hand+Forearm+Object, 224x224 pixels)
- Hand segmentation (Hand, 224x224 pixels)
- 3D Hand Pose annotation (21 keypoints)
- 3D Hand Shape annotation (MANO model)
- Camera calibration (Intrinsic+Distortion+Extrinsic)

In total there are 107,538 time steps available that were recorded by a multi-camera rig with 8 cameras, which results in 860,304 RGB images. Please note, RGB image and camera calibration is the only information that is always available.
For 102,480 time steps a predicted MANO fit is available, from which 9,055 were validated through manual inspection. Hand shape predictions were obtained using the FreiHAND method presented by Zimmermann et al. ICCV 2019.
Foreground segmentation was obtained exploiting the green screen background, which is available for 763,976 images (88.8%).
Hand Segmentation is derived from the Hand Shape annotation and consequently is available on the same frames (102,480*8 = 819,840 frames).
Similarly, 3D hand pose is derived from the hand shape annotation.
Be careful, when using non validated modalities! Their quality does differ substantially.

This dataset is an extended version of the FreiHand dataset.


Dataset Attributes

Background Randomization Time Sequences Multiple Views
The data was captured against a green-screen background, which allows for simple foreground detection and exchanging of the background. HanCo was captured in short video sequences. The dataset is captured with multiple calibrated and time synchronized cameras. In this video all cameras are iterated for a fixed time step. Three time steps are shown after each other.



Terms of use

This dataset is provided for research purposes only and without any warranty. Any commercial use is prohibited. If you use the dataset or parts of it in your research, you must cite the respective papers.

@InProceedings{Freihand2019,
  author    = {Christian Zimmermann, Duygu Ceylan, Jimei Yang, Bryan Russell, Max Argus and Thomas Brox},
  title     = {FreiHAND: A Dataset for Markerless Capture of Hand Pose and Shape from Single RGB Images},
  booktitle = {IEEE International Conference on Computer Vision (ICCV)},
  year = {2019},
  url          = "https://lmb.informatik.uni-freiburg.de/projects/freihand/"
}
@InProceedings{ZimmermannAB21,
  author    = {Christian Zimmermann, Max Argus, and Thomas Brox},
  title     = {Contrastive Representation Learning for Hand Shape Estimation}
  booktitle = {arxive},
  year = {2021},
  url          = "https://lmb.informatik.uni-freiburg.de/projects/hanco/"
}



Dataset

Tester

Since, the complete dataset is fairly large we provide a tester, which only contains a single sequence with all modalities in order to allow you to quickly see if this dataset is what you are looking for:

Download HanCo Tester (~60 MB)


Complete Data

Download HanCo RGB (~14 GB)
Download HanCo Foreground Masks (~3.7 GB)
Download HanCo Hand Masks (~2.7 GB)
Download HanCo 3D Hand Keypoints (79 MB)
Download HanCo Hand Shape Parameters (662 MB)
Download HanCo Camera Calibration and Meta Information (161 MB)

Images with newly inpainted backgrounds using different post processing strategies (see paper for details)
Download HanCo RGB with new background: Cut and Paste (19 GB)
Download HanCo RGB with new background: Homogenisation (7,3 GB)
Download HanCo RGB with new background: Sampling based image colorization (18 GB)
Download HanCo RGB with new background: Hallucination based image colorization (19 GB)




Contact

For questions about the dataset please contact Max Argus.