graviti
产品服务
解决方案
知识库
公开数据集
关于我们
Eye Gaze
Aesthetics
|...
许可协议: CC-BY-SA 4.0

Overview

Context

The main reason for making this dataset is the publication of the paper: Learning from Simulated and Unsupervised Images through Adversarial Training and the idea of the SimGAN. The dataset and kernels should make it easier to get started making SimGAN networks and testing them out and comparing them to other approaches like KNN, GAN, InfoGAN and the like.

Content

gaze.csv: A full table of values produced by the UnityEyes tool for every image in the gaze.h5 file

gaze.json: A json version of the CSV table (easier to read in pandas)

gaze.h5: The synthetic gazes from the UnityEyes tool

real_gaze.h5: The gaze images from MPII packed into a single hdf5

Acknowledgements

The synthetic images were generated with the windows version of UnityEyes http://www.cl.cam.ac.uk/research/rainbow/projects/unityeyes/tutorial.html

The real images were taken from https://www.mpi-inf.mpg.de/departments/computer-vision-and-multimodal-computing/research/gaze-based-human-computer-interaction/appearance-based-gaze-estimation-in-the-wild-mpiigaze/, which can be cited like this: Appearance-based Gaze Estimation in the Wild, X. Zhang, Y. Sugano, M. Fritz and A. Bulling, Proc. of the IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), June, p.4511-4520, (2015).

Inspiration

Enhancement:

  • One of the challenges (as covered in the paper) is enhancing the
    simulated images by using the real images. One possible approach is
    using the SimGAN which is implemented for reference in one of the
    notebooks. There are a number of other approaches (pix2pix, CycleGAN)
    which could have interesting results.

Gaze Detection:

  • The synthetic dataset has the gaze information since it was generated
    by UnityEyes with a predefined look-vector. The overview notebook
    covers what this vector means and how each component can be
    interpreted. It would be very useful to have a simple, quick network
    for automatically generating this look vector from an image
数据概要
数据格式
image,
数据量
429.452K
文件大小
571.23MB
发布方
4Quant
| 数据量 429.452K | 大小 571.23MB
Eye Gaze
Aesthetics
许可协议: CC-BY-SA 4.0

Overview

Context

The main reason for making this dataset is the publication of the paper: Learning from Simulated and Unsupervised Images through Adversarial Training and the idea of the SimGAN. The dataset and kernels should make it easier to get started making SimGAN networks and testing them out and comparing them to other approaches like KNN, GAN, InfoGAN and the like.

Content

gaze.csv: A full table of values produced by the UnityEyes tool for every image in the gaze.h5 file

gaze.json: A json version of the CSV table (easier to read in pandas)

gaze.h5: The synthetic gazes from the UnityEyes tool

real_gaze.h5: The gaze images from MPII packed into a single hdf5

Acknowledgements

The synthetic images were generated with the windows version of UnityEyes http://www.cl.cam.ac.uk/research/rainbow/projects/unityeyes/tutorial.html

The real images were taken from https://www.mpi-inf.mpg.de/departments/computer-vision-and-multimodal-computing/research/gaze-based-human-computer-interaction/appearance-based-gaze-estimation-in-the-wild-mpiigaze/, which can be cited like this: Appearance-based Gaze Estimation in the Wild, X. Zhang, Y. Sugano, M. Fritz and A. Bulling, Proc. of the IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), June, p.4511-4520, (2015).

Inspiration

Enhancement:

  • One of the challenges (as covered in the paper) is enhancing the
    simulated images by using the real images. One possible approach is
    using the SimGAN which is implemented for reference in one of the
    notebooks. There are a number of other approaches (pix2pix, CycleGAN)
    which could have interesting results.

Gaze Detection:

  • The synthetic dataset has the gaze information since it was generated
    by UnityEyes with a predefined look-vector. The overview notebook
    covers what this vector means and how each component can be
    interpreted. It would be very useful to have a simple, quick network
    for automatically generating this look vector from an image
0
立即开始构建AI
graviti
wechat-QR
长按保存识别二维码,关注Graviti公众号

Copyright@Graviti
沪ICP备19019574号
沪公网安备 31011002004865号