Pose Estimation
Pose Estimation 是计算机视觉领域的一项任务,旨在检测人物或物体的位置与姿态。该任务通过预测特定关键点(如手、头、肘等)的位置来实现对人体姿态的估计。Pose Estimation 在人机交互、运动分析、虚拟现实等领域具有重要应用价值。常见的基准测试包括 MPII Human Pose 数据集。
COCO test-dev
HRNet-W48+DARK
MPII Human Pose
PCT (swin-l, test set)
Leeds Sports Poses
OmniPose
OCHuman
ViTPose (ViTAE-G, GT bounding boxes)
CrowdPose
BUCTD-W48 (w/cond. input from PETR, and generative sampling)
COCO val2017
MogaNet-B (384x288)
AIC
MS COCO
OmniPose (WASPv2)
ITOP front-view
AdaPose
InLoc
GIM-DKM
ITOP top-view
DECA-D3
J-HMDB
SimpleBaseline + HANet
MPII Single Person
4xRSN-50
UPenn Action
OmniPose
SALSA
SubdivNet
300W (Full)
BRACE
HRNet fine-tuned on BRACE
COCO 2017 val
LOGO-CAP (Ours) HRNet-W48
DensePose-COCO
Parsing R-CNN + ResNext101
FLIC Elbows
Stacked Hourglass Networks
FLIC Wrists
Stacked Hourglass Networks
UAV-Human
AlphaPose
!(()&&!|*|*|
Nate
3DPW
ApolloCar3D
COCO minival
MSPN
KITTI 2015
GeoNet
MERL-RAV
SPIGA
MPII
OmniPose (WASPv2)
MS-COCO
UniHCP (finetune)
Pix3D
Mid-Level based