Pedestrian Attribute Recognition At Far Distance

Yubin Deng, Ping Luo, Chen Change Loy, Xiaoou Tang

Department of Informaiton Engineering, The Chinese University of Hong Kong

ACM MM 2014


The capability of recognizing pedestrian attributes, such as gender and clothing style, at far distance, is of practical interest in far-view video surveillance scenarios where face and body close-shots are hardly available. We make two contributions in this paper. First, we release a new pedestrian attribute dataset, which is by far the largest and most diverse of its kind. We show that the large-scale dataset facilitates the learning of robust attribute detectors with good generalization performance. Second, we present the benchmark performance by SVM-based method and propose an alternative approach that exploits context of neighboring pedestrian images for improved attribute inference.


Composition of PEdesTrian Attribute (PETA) dataset


The PETA dataset consists of 19000 images, with resolution ranging from 17-by-39 to 169-by-365 pixels. Those 19000 images include 8705 persons, each annotated with 61 binary and 4 multi-class attributes. The detail composition can be seen from the table below.

Datasets #Images Camera angle View point Illumination Resolution Scene
3DPeS 1012 high varying varying from 31x100 to 236 x 178 outdoor
CAVIAR4REID 1220 ground varying low from 17x39 to 72x141 outdoor
CUHK 4563 high varying varying 80x160 indoor
GRID 1275 varying frontal & back low from 29x67 to 169x365 indoor
i-LIDS 477 medium back high from 32x76 to 115x294 outdoor
MIT 888 ground back high 64x128 outdoor
PRID 1134 high profile low 64x128 outdoor
SARC3D 200 medium varying varying from 54x187 to 150x307 outdoor
TownCentre 6967 medium varying medium from 44x109 to 148x332 outdoor
VIPeR 1264 ground varying varying 48x128 outdoor
Total = PETA 19000 varting varying varying varying varying


Sample Images in PETA Dataset



Y. Deng, P. Luo, C. C. Loy, X. Tang, "Pedestrian attribute recognition at far distance," in Proceedings of ACM Multimedia (ACM MM), 2014  [PDF]



This dataset is intended for research purposes only and as such cannot be used commercially. In addition, reference must be made to the aforementioned publications when this dataset is used in any academic and research reports.

PETA dataset (on Dropbox)    PETA dataset (on FTP)