MARS Attributes and DukeMTMC-VID Attributes

Information

We labeled 20 attributes for each pedestrain sequence in MARS and 18 attributes in DukeMTMC-VID, the definition of the attributes is fully described in the following section. The detailed information of MARS and DukeMTMC-VID can be found at MARS and DukeMTMC-VID, for the reason that in the test-set of MARS, the trackets which belong to the ID "0000" and "00-1" is junk images and distractors, which can't be used to do perdestrian attribute recognition, so these trackets is excluded from the MARS-attribute dataset.

Datasets download

Baiduyun
password: beda

Attribute definition of the MARS-Attribute dataset

For the Action attribute, the value 0~4 means the motion of the pedestrain is [walking, running, riding, staying, varies]
For the Angle attribute, the value 0~5 means the pose of the pedestrain is [frontal, lateral-frontal, lateral, lateral-back, back, varies]
For the Tops.c attribute, the value 0~8 means the color of tops is [black, purple, green, blue, gray, white, yellow, red, complex]
For the Bottoms.c attribute, the value 0~9 means the color of bottoms is [white, purple, black, green, gray, pink, yellow, blue, brown, complex]
For the Age attribute, the value 0~3 means the age is [young, teenager, adult, old]
Other attributes are binary attributes.

Attribute definition of the DukeMTMC-VID Attribute dataset

For the Action attribute, the value 0~4 means the motion of the pedestrain is [walking, running, riding, staying, varies]
For the Angle attribute, the value 0~5 means the pose of the pedestrain is [frontal, lateral-frontal, lateral, lateral-back, back, varies]
For the Tops.c attribute, the value 0~8 means the color of tops is [black, white, red, purple, gray, blue, green, brown, complex]
For the Bottoms.c attribute, the value 0~7 means the color of bottoms is [black, white, red, gray, blue, green, brown, complex]
Other attributes are binary attributes.

Publications

ZhiyuanChen, Annan Li and Yunhong Wang: A Temporal Attentive Approach for Video-Based Pedestrian Attribute Recognition, Chinese Conference on Pattern Recognition and Computer Vision (PRCV), Xi'an, China, 2019.