/HumanMesh

[AMILab] 2021 Summer Research

Primary LanguagePython

[AMILab] HumanMesh

Efficient Human behavior understanding_3d pose estimate

WORKS

  • 논문 하루에 하나씩

목표

논문을 공부하고 HMR을 기반으로하여, 백본 네트워크를 본인의 아이디어로 아키텍처를 재구성하고 학습시켜 성능을 유지하면서 속도 올리기(RealTime)

  • Single Image 기반 Human Mesh Recovery에서 더 도전적으로 [VIBE, CVPR2020]
  • Bounding Box를 detection하면서 동시에 human mesh recovery가 가능한 형태로 Mask R-CNN이나 Yolo등의 Two/One stage방법등으로 확장 시도

References

3D human mesh Recovery

  • SMPL:A Skinned Multi-Person Linear Model, ACM Trans. Graphics (Proc. SIGGRAPH Asia), 2015
  • Keep it {SMPL}: Automatic Estimation of {3D} Human Pose and Shape from a Single Image, ECCV 2016
  • End-to-end Recovery of Human Shape and Pose, CVPR 2018 참고
  • Learning to Reconstruct 3D Human Pose and Shape via Model-fitting in the Loop, ICCV 2019 참고
  • VIBE: Video Inference for Human Body Pose and Shape Estimation, CVPR 2020
  • End-to-End Human Pose and Mesh Reconstruction with Transformers, CVPR 2021 참고

Models and Light-weight models

  • Mask R-CNN, ICCV 2017
  • Focal Loss for Dense Object Detection, ICCV 2017(RetinaNet)
  • YOLACT:Real-time Instance Segmentation, ICCV 2019
  • MobileNets: Efficient Convolutional Neural Neetworks for Mobile Vision Applications, arXiv 2017
  • CONVOLUTIONAL NEURAL NETWORKS WITH LOWRANK REGULARIZATION, ICLR 2016