American Sign Language Image Dat

American Sign Language Image Dataset (ASLID)

Srujana Gattupalli, Amir Ghaderi
Vision-Learning-Mining Laborartory
The Universiy of Texas at Arlington, TX, USA

Overview
We present the American Sign Language Image Dataset (ASLID) with images extracted from Gallaudet Dictionary videos[1] and American Sign Language Lexicon Video Dataset(ASLLVD)[2] with annotations for upper body joint locations. The annotation file contains image video and frame name and 7 upper body joint co-ordinates, namely left hand, left elbow, left shoulder, head, right hand, right shoulder, right wrist. We also provide a tool to visualize the annotations. For the purpose of Upper body pose estimation experiments we have segregated the training and test set.

Training Images
Our training set comprises of 809 ASLID images(240 X 352) from different signs from six different native ASL signers from Gallaudet Dictionary videos[1] and annotations for them. We have not provided images here due to copyright issues and can be requested by emailing authors. The annotations tool to display annotations and tool to extract image given a video and frame number are provided in section below.

Test Images
The test set we have used 479 ASLID images (480 X 640) from two native ASL signers from ASLLVD videos[2]. Note:The frame numbers used in our annotations start from 1. The annotations shown in video frames start from 0.

Images can be downloaded from the following link: ASLID_Testset

Annotations

Annotations Description-

Annotaions for train and test images are stored in a matlab structure “train_data.mat” and “test_data.mat” respectively. They have these fields:

image_name : the cell which contain the name of images.

left_wrist_x : x coordinate of the left wrist

left_wrist_y : y coordinate of the left wrist

left_elbow_x : x coordinate of the left elbow

left_elbow_y : y coordinate of the left elbow

left_shoulder_x : x coordinate of the left shoulder

left_shoulder_y : y coordinate of the left shoulder

head_x : x coordinate of the head

head_y : y coordinate of the head

right_shoulder_x : x coordinate of the right shoulder

right_shoulder_y : y coordinate of the right shoulder

right_elbow_x : x coordinate of the right elbow

right_elbow_y : y coordinate of the right elbow

right_wrist_x : x coordinate of the right wrist
Annotations link - Annotations for ASLID training can be downloaded with the following link: train_data.mat

Annotations link - Annotation for ASLID testing can be downloaded with the following link: test_data.mat
Visualizing Annotations:
- The following Matlab code allows you select image to visualize annotations: showAnnot.m
Example ASLID annotation visualizations:

example example example example

Variation in range of training poses is shown in the ground truth scatter plot below:

plot

Variation in range of test set poses is shown in the ground truth scatter plot below:

plot

Citation:
If you use our dataset for your research work, please Cite our Paper:
"Evaluation of Deep Learning based Pose Estimation for Sign Language", Srujana Gattupalli, Amir Ghaderi, Vassilis Athitsos, International Conference on PErvasive Technologies Related to Assistive Environments, PETRA, Corfu Island Greece, June 2016.

References:
[1]C. Valli, editor. The Gallaudet Dictionary of American Sign Language. Gallaudet U. Press, Washington, DC, 2006.
[2]http://www.bu.edu/av/asllrp/dai-asllvd.html

Thank You for Visiting our Webpage!