Invariant Teacher and Equivariant Student for Unsupervised 3D Human Pose Estimation (2020-12-17T00:00:00.000000Z)

TL;DR

A novel method based on teacher-student learning framework for 3D human pose estimation without any 3D annotation or side information is proposed and reduces the 3D joint prediction error compared to state-of-the-art unsupervised methods and also outperforms many weakly- supervised methods that use side information on Human3.6M.

Abstract

We propose a novel method based on teacher-student learning framework for 3D human pose estimation without any 3D annotation or side information. To solve this unsupervised-learning problem, the teacher network adopts pose-dictionary-based modeling for regularization to estimate a physically plausible 3D pose. To handle the decomposition ambiguity in the teacher network, we propose a cycle-consistent architecture promoting a 3D rotation-invariant property to train the teacher network. To further improve the estimation accuracy, the student network adopts a novel graph convolution network for flexibility to directly estimate the 3D coordinates. Another cycle-consistent architecture promoting 3D rotation-equivariant property is adopted to exploit geometry consistency, together with knowledge distillation from the teacher network to improve the pose estimation performance. We conduct extensive experiments on Human3.6M and MPI-INF-3DHP. Our method reduces the 3D joint prediction error by 11.4% compared to state-of-the-art unsupervised methods and also outperforms many weakly-supervised methods that use side information on Human3.6M. Code will be available at https://github.com/sjtuxcx/ITES.

Authors

Maosen Li

3 papers

Siheng Chen

3 papers

Ya Zhang

3 papers

TL;DR

Abstract

Authors

References40 items

Geometry-Driven Self-Supervised Method for 3D Human Pose Estimation

Kinematic-Structure-Preserved Representation for Unsupervised 3D Human Pose Estimation

Dynamic Multiscale Graph Neural Networks for 3D Skeleton Based Human Motion Prediction

Learning to Reconstruct 3D Human Pose and Shape via Model-Fitting in the Loop

C3DPO: Canonical 3D Pose Networks for Non-Rigid Structure From Motion

Distill Knowledge From NRSfM for Weakly Supervised 3D Pose Learning

Optimizing Network Structure for 3D Human Pose Estimation

IGE-Net: Inverse Graphics Energy Networks for Human Pose Estimation and Single-View Reconstruction

Actional-Structural Graph Convolutional Networks for Skeleton-Based Action Recognition

Unsupervised 3D Pose Estimation With Geometric Self-Supervision

Semantic Graph Convolutional Networks for 3D Human Pose Regression

Self-Supervised Learning of 3D Human Pose Using Multi-View Geometry

RepNet: Weakly Supervised Training of an Adversarial Reprojection Network for 3D Human Pose Estimation

3D Human Pose Estimation in Video With Temporal Convolutions and Semi-Supervised Training

Can 3D Pose be Learned from 2D Projections Alone?

Image Collection Pop-up: 3D Reconstruction and Clustering of Rigid and Non-rigid Categories

Ordinal Depth Supervision for 3D Human Pose Estimation

Unsupervised Adversarial Learning of 3D Human Pose from 2D Joint Locations

Learning Monocular 3D Human Pose Estimation from Multi-view Images

End-to-End Recovery of Human Shape and Pose

Cascaded Pyramid Network for Multi-person Pose Estimation

Rocket Launching: A Universal and Efficient Framework for Training Well-performing Light Net

Adversarial Inverse Graphics Networks: Learning 2D-to-3D Lifting and Image-to-Image Translation from Unpaired Supervision

A Simple Yet Effective Baseline for 3d Human Pose Estimation

Towards 3D Human Pose Estimation in the Wild: A Weakly-Supervised Approach

3D Human Pose Estimation = 2D Pose Estimation + Matching

Monocular 3D Human Pose Estimation in the Wild Using Improved CNN Supervision

Semi-Supervised Classification with Graph Convolutional Networks

Prior-Less Compressible Structure from Motion

Single Image 3D Interpreter Network

Sparseness Meets Deepness: 3D Human Pose Estimation from Monocular Video

Grouping-Based Low-Rank Trajectory Completion and 3D Reconstruction

Human3.6M: Large Scale Datasets and Predictive Methods for 3D Human Sensing in Natural Environments

Complex Non-rigid Motion 3D Reconstruction by Union of Subspaces

Reconstructing 3D Human Pose from 2D Image Landmarks

A Simple Prior-Free Method for Non-rigid Structure-from-Motion Factorization

Trajectory Space: A Dual Representation for Nonrigid Structure from Motion

Nonrigid Structure from Motion in Trajectory Space

Recovering non-rigid 3D shape from image streams

Large-Scale Machine Learning with Stochastic Gradient Descent

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names