ReachPredict3D

Reach Prediction

Here we outline our software process to predict user-chosen appendage locations with marker-less pose estimation using DeepLabCut ( https://github.com/DeepLabCut/DeepLabCut ) in 3-D. We choose 27 unique positions to track across all rats in our experiments. These positions form the basis for our pose estimation during reaching behavior. We then reconstruct the individual camera 2-D scene spaces into a pre-calibrated euclidean 3-D space using Direct Linear Transformations. These 3-D predictions and their associated confidence intervals are then saved into the NWB format, per session.

More information about the ReachPredict3D pipeline can be found on our tutorial!

Prediction

3-D Reach Reconstruction Main Functions

Above are main functions to import, export, and load the various utilities necessary for 3-D reconstruction of our inferred points. More detail is paid to individual utility functions in the preprocessing software documentation.