level: nature neuroscience author:Alexander Mathis  1,2, Pranav Mamidanna1 , Kevin M. Cury3 , Taiga Abe3 , Venkatesh N. Murthy  2 , Mackenzie Weygandt Mathis  1,4,8* and Matthias Bethge1,5,6,7,8 date: 2018 keyword:

• quantifying behavior; pose estimation;

Mathis A, Mamidanna P, Cury K M, et al. DeepLabCut: markerless pose estimation of user-defined body parts with deep learning[J]. Nature neuroscience, 2018, 21(9): 1281-1289. cited by 401;

Paper: DeepLabCut

Summary

1. present an efficient method for markerless pose estimation based on transfer learning with deep neural networks;
2. by labeling only a few hundred frames, one can train tailored, robust feature detectors that are capable of localizing a variety of experimentally relevant body part;

Proble Statement

• inhomogeneous illumination; transparent side walls that appear dark; shadows around the mouse from over head lighting; distortions due to a wide-angle lens; the frequent crossing of the mouse over the odor trail; the common occurrence of rewards directly in front of its snout;

previous work:

• physical markers;
• fit skeleton or active contour models13-17;
• training regressors based on various computationally derived features to track particular body parts in a supervised way;

Methods

• Problem Formulation:

• system overview:

Notes 去加强了解

• Deepcut
• https://github.com/DeepLabCut/DeepLabCut 学习使用