README.md

1.Dataset Tree:

 ntu_dataset/senz3d_dataset
   |__ images-full
   |__ images-part
   |__ keypoint
   |__ skeleton
   |__ boundary
   |__ triangle
   |__ test.part.lst
   |__ test.full.lst
   |__ train.part.lst
   |__ train.full.lst

2.Description:

File/Folder Name	Description	Number/Lines	Others
ntu_dataset	NTU Hand Gesture Dataset, 10 images for each gesture with 10 kinds of gestures collected from 10 people.	1000	-
senz3d_dataset	Senz3D Gesture Dataset, 30 images for each gesture with 11 kinds of gestures collected from 4 people.	1320	-
image-full	All of the images in the public dataset.	1000(ntu)/1320(senz3d)	color
image-part	Part of the images in public dataset, including the images that gesture can be detected via OpenPose.	647(ntu)/494(senz3d)	color, resized
keypoint	Key-points map of hand gesture.	647(ntu)/494(senz3d)	binary
skeleton	Skeleton map of hand gesture.	647(ntu)/494(senz3d)	binary
boundary	Rough boundary map of hand gesture.	0(ntu)/1320(senz3d)	binary
triangle	Rough triangle map of hand gesture.	1000(ntu)/1320(senz3d)	binary
test.part.lst	Testing list with image-part	9600(ntu)/12800(senz3d)	-
test.full.lst	Testing list with image-full	13500(ntu)/16500(senz3d)	-
train.part.lst	Training list with image-part	84636(ntu)/135504(senz3d)	-
train.full.lst	Training list with image-full	44100(ntu)/138864(senz3d)	-

• Notes:

  • Keypoints, Skeleton: based-on the public project OpenPose (Notes: OpenPose can't detect all hand gestures in the two datasets).
  • Boundary, Triangle: manually annotated.

• Additional:

In NTU Dataset, it can make the model confusing that what the real direction of hand gesture is, as shown in the following table. We ask the users to provide some more simple information to confirm that.

Image	Triangle	Modified Triangle

3.Download

You can download the dataset from Google Drive.