Related Experiment Video
Updated: Oct 25, 2025
07:52
Automated Rat Single-Pellet Reaching with 3-Dimensional Reconstruction of Paw and Digit Trajectories
Published on: July 10, 2019
14.6K
LiftPose3D, a deep learning-based approach for transforming two-dimensional to three-dimensional poses in laboratory
Adam Gosztolai1, Semih Günel2,3, Victor Lobato-Ríos4
1Neuroengineering Laboratory, Brain Mind Institute & Interfaculty Institute of Bioengineering, EPFL, Lausanne, Switzerland. adam.gosztolai@epfl.ch.
Nature Methods
|August 6, 2021
Summary
LiftPose3D reconstructs 3D animal poses from a single 2D camera view, simplifying kinematic studies. This markerless method eliminates the need for multiple cameras and complex calibration, making 3D pose estimation more accessible.
Area of Science:
- Animal behavior analysis
- Biomechanical research
- Computational neuroscience
Background:
- Markerless three-dimensional (3D) pose estimation is crucial for animal kinematic studies.
- Current multi-view triangulation methods require multiple cameras and complex calibration, limiting their use.
- There is a need for accessible and robust 3D pose estimation techniques.
Purpose of the Study:
- To introduce LiftPose3D, a novel deep network-based method for reconstructing 3D poses from a single 2D camera view.
- To demonstrate the versatility and accuracy of LiftPose3D across various animal models and behaviors.
- To overcome the limitations of traditional 3D pose estimation methods.
Main Methods:
- Developed a deep network architecture for lifting 2D pose estimates to 3D.
- Utilized a single camera input for pose reconstruction.
- Validated the method on diverse datasets including flies, mice, rats, and macaques.
Main Results:
- LiftPose3D accurately reconstructs 3D poses from monocular 2D data.
- The method performs well across different species and behaviors, including occluded poses.
- Achieved high-quality 3D pose estimation without complex multi-camera setups or calibration.
Conclusions:
- LiftPose3D offers a simplified and effective solution for markerless 3D animal pose estimation.
- The framework enhances the accessibility of 3D kinematic analysis in laboratory settings.
- Enables advanced behavioral studies previously hindered by technical constraints.
