Initializing labels

Initialized labeling

One simple way to speed up labeling is to initialize the labels with predictions from a network trained on a few frames. Initializing keypoint predictions is easy with Label3D.loadFrom3D (Note: This is the same function used to load predictions in View3D). Just load frames into Label3D as you would normally, then initialize the keypoints with loadFrom3D.

labelGui = Label3D(params, videos, skeleton);

% Load dannce predictions from a file. pts3d.pred is nFrames x 3 x nKeypoints 
pts3d = load('/path/to/dannce/predictions.mat')
labelGui.loadFrom3D(pts3d.pred) 

These loaded keypoints will appear as initialized (gray) in the status indicator as opposed to labeled (white) or unlabeled (black). You can manually adjust the keypoints as you would when labeling without initialized labels. When you are happy with the keypoints for a frame, mark all keypoints from a single frame as labeled rather than initialized by pressing the letter l.