Introduction
We conduct the comprehensive study of sensing human beauty via single/multi-modality cues. The collected dataset contains three modalities, i.e., face, dressing and voice. The attractiveness scores are given by k-wise preferences from participants. All modalities are labeled with extensive attributes by Amazon Mechanical Turk due to its large amount. Both visual and vocal features as well as labeled attributes are collectively utilized to build computational models for estimating female beauty score. The paper can be found here.
Dataset
This page has links for downloading the M2B dataset, which consists of facial, dressing images and audio files of 1240 females. In this paper, Mongoloid females such as the ones with Chinese, Korean and Japanese origins represent the Eastern group, whereas Caucasian females, who are descended from Angles, Celtic, Latin and Germanic people, represent the Western group [1]. We have removed the ambiguous cases such as Arabic and Eurasian females.
[1] BRINTON, D. 1890. Races and peoples: lectures on the science of ethnography. N.D.C. Hodges.
The raw data are provided along with the extracted features as mat files. In total there are 5 files that need to be downloaded, 3 of which are raw data files consisting of (i) facial images; (ii) dressing images; (iii) audio files. The other two files are the Matlab mat files that give you the beauty scores and extracted features.
Downloads
1. Facial images (Download)
2. Dressing images (Download)
3. Audio files (Download)
4. Facial Landmark points (Download)
4. Beauty scores (Download)
5. Extracted features (Download)
Citation
Please cite the following papers if you want to use this dataset in your research:
[1] Tam V. Nguyen, Si Liu, Bingbing Ni, Jun Tan, Yong Rui, Shuicheng Yan: Sense beauty via face, dressing, and/or voice. ACM Multimedia 2012: 239-248
[2] Tam V. Nguyen, Si Liu, Bingbing Ni, Jun Tan, Yong Rui, Shuicheng Yan: Towards decrypting attractiveness via multi-modality cues. TOMCCAP 9(4): 28 (2013)
Contact
tamnguyen@nus.edu.sg