Elife https://doi.org/10.7554/eLife.47994 (2019).
Elife https://doi.org/10.7554/eLife.48571 (2019).
An in-depth study of animal behavior at the individual level requires analysis of the animal’s pose. Recent deep-learning-based approaches include DeepLabCut and LEAP. Graving et al. developed an optimized deep learning architecture called Stacked DenseNet to surpass DeepLabCut and LEAP in inference speed and accuracy. The researchers applied their DeepPoseKit framework to datasets of fruit flies, locusts and zebras. While DeepPoseKit is optimized to analyze posture in 2D in freely moving animals, Günel et al. developed a deep learning pipeline to monitor the pose of tethered fruit flies in 3D while they walked on a ball. DeepFly3D uses images from seven cameras to identify 38 landmarks on the animal’s body and to generate a representation of a fruit fly’s pose in 3D. The researchers used their pipeline to perform a behavioral classification task in control and optogenetically manipulated fruit flies.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Vogt, N. Pose estimation with deep learning. Nat Methods 16, 1205 (2019). https://doi.org/10.1038/s41592-019-0678-2
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41592-019-0678-2