Overview

We offer a benchmark suite together with an evaluation server, such that authors can upload their results and get a ranking. We offer a dataset that contains more than 25,000 pictures, including 15,403 images for training set, 5,000 images for validation set and 5,000 images for testing set. If you would like to submit your results, please follow the instructions on our submission page.

Note: We only display results with relatively detailed descriptions.

Multi-Human Parsing

Metrics

We use two human-centric metrics for multi-human parsing evaluation, which are initially reported by the MHP v1.0 paper. The two metrics are Average Precision based on part (AP^p) (%) and Percentage of Correctly parsed semantic Parts (PCP) (%).

Ranking

All teams with successful submissions have a placeholder in the leaderboard, and the results of all teams will be released on 10 June. The winner of the challenge is the team with maximal numbers of top-1 ranking among all the five metrics (one in each columns). Ties are broken by the score of AP^p_0.5.

Method	APp0.5	APpvol	PCP0.5	APP0.5(Inter20%)	APP0.5(Inter10%)	Abbreviation	Submit Time
Baseline	25.14	41.78	32.25	18.61	14.88	Abbreviation Contributors Description "Understanding Humans in Crowded Scenes: Deep Nested Adversarial Learning and A New Benchmark for Multi-Human Parsing", Jian Zhao*, Jianshu Li*, Yu Cheng*, Li Zhou, Terence Sim, Shuicheng Yan, Jiashi Feng; arXiv:1804.03287 (* indicates equal contribution) Nested Adversarial Network(NAN) solves multi-human parsing problem by simultaneously performing 1) semantic saliency prediction, 2) instance-agnostic parsing and 3) instance-aware clustering.	2018-04-12 20:00:00
SKK-2	9.98	31.40	21.03	3.68	1.93	Abbreviation Contributors Description SKK-2: Bin Wen(Beijing Forestry University), Weiqi Zhang(Beijing Forestry University) Our baseline model are Mask-RCNN and deeplab.	2018-05-09 14:04:00
S-LAB	31.47	40.71	38.27	17.86	10.33	Abbreviation Contributors Description S-LAB: Cai Xiaofan, Huang Shaofei, Si Liu (Institute of Information Engineering, Chinese Academy of Sciences) MASK-RCNN & PSPNET+deeplabv3+ for parsing	2018-06-11 07:26:00
BJTU_UIUC	33.34	42.25	41.82	12.45	6.53	Abbreviation Contributors Description BJTU_UIUC: PSPse, Tao Ruan(BJTU)*, Ting Liu(BJTU)*, Yunchao Wei(UIUC), Jie Zhao(BJTU), Shikui Wei(BJTU), Honghui Shi(IBM), Yao Zhao(BJTU), Thomas Huang(UIUC)(* means equal contribution) Upon our PSPse network for single human parsing (as detailed in Track 1), we propose to leverage Mask RCNN + PSPse for conducting multiple human parsing. We firstly employ Mask RCNN to infer the mask of each human instance. Second, each inferred instance is then cropped and fed into two PSPse networks, which are learned by different instance-level cropping policies, for predicting part-level segmentation. By augmenting the parsing predictions from global view, we finally obtain our multiple human parsing results. Ref: [1] Pyramid Scene Parsing Network. CVPR 2017	2018-06-10 04:56:00
UTP	16.44	34.38	28.39	5.51	2.69	Abbreviation Contributors Description TBD Detection the human by a detection network, and then do segmentation on body area. Training data: mph train data, pretrained model by image-net	2018-06-10 15:33:00