Retraining mis-labeled points

deeplabcut
#1

Hi,

I labeled 22 points in some frames and do the training. When I use the trained results to analyze a new video, 20 of points work perfectly well and 2 of the points are wrong. To improve the accuracy of the two points, can I only label the two points in the new video to retrain the model? Or we have to label all the 22 points?

Thanks,
Yu

0 Likes

#2

Deeplabcut has a build in function to do this. see here: https://github.com/AlexEMG/DeepLabCut/blob/master/docs/functionDetails.md#j-refinement-extract-outlier-frames

1 Like

#3

Thanks. That solved my problem.

0 Likes