Core Concepts
Recovering human pose and camera trajectories in the world coordinate system using WHAC framework.
Abstract
The study introduces WHAC, a novel framework for estimating expressive human models and camera poses in the world coordinate system. It leverages insights from human motions, camera-frame estimation methods, and visual odometry to achieve accurate scale recovery. Additionally, a synthetic dataset, WHAC-A-Mole, is presented to facilitate benchmarking. The study demonstrates superior performance on both standard benchmarks and the newly established WHAC-A-Mole dataset.
Abstract:
Introduces WHAC framework for recovering human pose and camera trajectories in the world coordinate system.
Presents a synthetic dataset, WHAC-A-Mole, for benchmarking.
Demonstrates superior performance on standard benchmarks.
Introduction:
Highlights the importance of estimating 3D trajectories in the world coordinate system.
Discusses challenges with existing EHPS methods focusing on parametric human models in the camera coordinate system.
Introduces the need for accurate scale recovery without relying on traditional optimization techniques.
Methodology:
Explains how WHAC synergizes human-camera interactions to estimate expressive human models and camera movements.
Details the process of recovering scaled human and camera trajectories using MotionVelocimeter.
Describes the generation of camera trajectories based on character locations and facing directions.
Experiments:
Evaluates WHAC performance on both camera-frame and world-grounded benchmarks.
Compares results with existing state-of-the-art methods showcasing superior performance in scale recovery and trajectory accuracy.
Conclusion:
Concludes that WHAC achieves state-of-the-art results in recovering human pose and camera trajectories.
Acknowledges limitations related to close interactions and occlusions not addressed by WHAC.
Raises concerns about potential negative societal impacts related to surveillance applications.
Stats
"Estimating 3D trajectories in the world coordinate system from monocular video is challenging."
"Camera-frame SMPL-X estimation methods recover absolute human depth."
"WHAC pioneers whole-body estimation in a world-grounded context."
Quotes
"Our approach is founded on two key observations."
"WHAC pioneers whole-body, optimization-free estimation."