CrowdHuman: A Benchmark for Detecting Human in a Crowd

By Shuai Shao, Zijian Zhao, Boxun Li, Tete Xiao, Gang Yu, Xiangyu Zhang. 2018

866 citations recorded by the source corpus.

Human detection has witnessed impressive progress in recent years. However, the occlusion issue of detecting human in highly crowded environments is far from solved. To make matters worse, crowd scenarios are still under-represented in current human detection benchmarks. In this paper, we introduce a new dataset, called CrowdHuman, to better evaluate detectors in crowd scenarios. The CrowdHuman dataset is large, rich-annotated and contains high diversity. There are a total of $470K$ human instances from the train and validation subsets, and $~22.6$ persons per image, with various kinds of occlusions in the dataset. Each human instance is annotated with a head bounding-box, human visible-region bounding-box and human full-body bounding-box. Baseline performance of state-of-the-art detection frameworks on CrowdHuman is presented. The cross-dataset generalization results of CrowdHuman dataset demonstrate state-of-the-art performance on previous dataset including Caltech-USA, CityPersons, and Brainwash without bells and whistles. We hope our dataset will serve as a solid baseline and help promote future research in human detection tasks.

Subfields: Artificial Intelligence, Computer Science Applications, Computer Vision and Pattern Recognition, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Science Applications

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

Citations in the corpus

iShape: A First Step Towards Irregular Shape Instance Segmentation (2021, neutral): “Instead, in the scenario of human detection and segmentation, the OC human [19] and the Crowd Human [28] introduce datasets with larger MaxIoU.”
Occluded Video Instance Segmentation: A Benchmark (2021, background): “One interpretation is that adjusting the NMS threshold is more important for tasks that require detecting the amodal bounding boxes (additionally containing the occluded invisible parts), such as the full-body bounding boxes in crowded pedestrian detection datasets (Shao et al., 2018; Zhang et al., ”
Deep Learning-based Human Pose Estimation: A Survey (2020, background): “, CrowdHuman [288], to improve person detection under severe occlusion.”
Pedestrian Detection: The Elephant In The Room (2020, methodology): “Interestingly, Wider Pedestrian [1] appears to be slightly better source of pre-training than CrowdHuman [31] since it contains more images of autonomous driving scenes.”
TransTrack: Multiple-Object Tracking with Transformer (2020, methodology): “Following [45], we pre-train our network on CrowdHu-man [31].”
StrongSORT: Make DeepSORT Great Again (2022, methodology): “We train the detector on the CrowdHuman dataset [41] and MOT17 half training set for ablation following [66, 63, 45, 56, 61].”
Efficient DETR: Improving End-to-End Object Detector with Dense Prior (2021, neutral): “We evaluate our method on the CrowdHuman dataset [29] to verify the robustness of Efficient DETR in crowded scenes.”
CrowdPose: Efficient Crowded Scenes Pose Estimation and a New Benchmark (2018, background): “However, the two-step approaches highly depend on human detection results, and it fails in crowded scenes [26].”
Simple Multi-dataset Detection (2021, methodology): “Specifically, we test on VIPER [32], Cityscapes [3], ScanNet [4], WildDash [44], KITTI [11], Pascal VOC [6], and CrowdHuman [34].”
Omni-DETR: Omni-Supervised Object Detection with Transformers (2022, neutral): “On CrowdHuman, for the cost of about 330 hours OSOD improves the strong SSOD baseline mAP by ∼ 4%.”
OTA: Optimal Transport Assignment for Object Detection (2021, methodology): “Hence to demonstrate OTA’s advantage on handling ambiguous anchors, it is necessary to conduct experiments on a crowded dataset – Crowdhuman [35].”
FCOS: A Simple and Strong Anchor-Free Object Detector (2020, methodology): “We also conduct experiments on the highly crowded dataset CrowdHuman [34].”
Rethinking the Competition Between Detection and ReID in Multiobject Tracking (2020, methodology): “Moreover, we further add CrowdHuman [36] images into training, and observe that it brings gains of 2.7 point on Impact of ID Loss Weight.”
BoT-SORT: Robust Associations Multi-Pedestrian Tracking (2022, methodology): “To avoid the possible inﬂuence caused by the detector, we used Byte-Track’s YOLOX-X MOT17 ablation study weights which were trained on CrowdHuman [38] and MOT17 ﬁrst half of the train sequences.”
Bounding Box Regression With Uncertainty for Accurate Object Detection (2018, methodology): “Large scale object detection datasets like ImageNet [6], MS-COCO [35] and CrowdHuman [47] try to define the ground truth bounding boxes as clear as possible.”
Adaptive NMS: Refining Pedestrian Detection in a Crowd (2019, methodology): “Recently, CrowdHuman [36] has been released to specifically target to the crowd issue in the human detection task.”
Monocular, One-stage, Regression of Multiple 3D People (2020, methodology): “CMU Panoptic.”
TrackFormer: Multi-Object Tracking with Transformers (2021, methodology): “We indicated additional training Data: CH=CrowdHuman [45], PD=Parallel Domain [50], 6M=6 tracking datasets as in [66], JTA [14], M=Market1501 [67] and C=CUHK03 [27].”
DiffusionDet: Diffusion Model for Object Detection (2022, methodology): “For modern object detection benchmarks [20,34,57,80], the number of instances of interest typically varies across images.”