##########################
### EgoDexter Dataset ###
##########################

----------------------------------------------------------------------------------------------------------------------------------------------
|  Terms of use:                                                                                                                             |
|  The provided dataset is intended for research purposes only and any use of it for non-scientific and/or commercial means is not allowed.  |
|  This includes publishing any scientific results obtained with our data in non-scientific literature, such as tabloid press.               |
|  If you use this dataset, you are required to cite the following paper:                                                                    |
|                                                                                                                                            |
|  Mueller F, Mehta D, Sotnychenko O, Sridhar S, Casas D, Theobalt C.                                                                        |
|  Real-time Hand Tracking under Occlusion from an Egocentric RGB-D Sensor.                                                                  |
|  Proc. of the IEEE International Conference on Computer Vision 2017.                                                                       |
|                                                                                                                                            |
|  Refer to the license (license.txt) distributed with the data.                                                                             |
|                                                                                                                                            |
----------------------------------------------------------------------------------------------------------------------------------------------

This benchmark dataset contains 4 RGB-D sequences recorded from a shoulder-mounted Intel RealSense SR300 camera (resolution 640x480 pixels for both, color and depth).
The data features 4 different actors (2 female), diverse challenging interaction with objects as well as varying cluttered backgrounds.
Fingertip positions in 3D (for visible fingertips) were manually annotated for TODO out of TODO frames. 

### Directory Structure ###
* data/
	* <SEQUENCE_NAME>/
		* color/
		* color-on-depth/
		* depth/
		* annotation/
* preview/
* CameraCalibration_640x480.txt
* license.txt
* README.txt (this file)
* Results_MuellerICCV2017.mat

### Provided Data ###		
For every frame, the following data is provided:
- depth: 			16-bit depth image (in mm)
- color_on_depth: 	24-bit color image mapped to the depth view 
					(i.e., the intrinsics and extrinsics correspond to the depth camera and only valid depth pixels have an assigned color)
- color: 			24-bit color image

### Annotation Format ###
Every line corresponds to one frame. The order of fingertips is THUMB, INDEX, MIDDLE, RING, LITTLE. 
- annotation.txt: 	This file contains the 2D annotations (F1_u, F1_v; F2_u,...). If a frame is not annotated, the 2D annotations are all -1. 
					If a fingertip is occluded, the annotator clicked on a background depth pixel, i.e. an occluded fingertip can only be identified by looking up the corresponding depth value.
- annotation.txt_3D.txt: 	This file contains the 3D annotations obtained from backprojecting the 2D annotation (F1_x, F1_y, F1_z; F2_x,...).
							If a frame is not annotated, the 3D annotations are all 0. If a fingertip is occluded, the respective x,y,z values are set to 0. 
							This means, in contrast to the 2D annotation file, occluded fingertips can be identified from the annotation file alone.
							
### Our Results (Mueller et al, ICCV 2017) ###
The file Results_MuellerICCV2017.mat contains the average 3D Euclidean errors and standard deviations for each sequence (variables Error_<SEQUENCE_NAME> and StdDev_<SEQUENCE_NAME>).
Furthermore, the variable PercentageBelowThresh provides the percentage of frames (row 2) below a certain threshold (row 1).