Recent advances in computer vision have enabled
the development of automated animal behavior observation
tools. Despite their generally encouraging
performance, multi-animal tracking tools still
face challenges, particularly with “body swapping”
– failure to maintain identities across time. Here we
present DIPLOMAT, multi-animal tracking software
tool that greatly reduces identity assignment errors
by introducing a combination of (i) an automated
pose estimate post-processing algorithm (“Track”)
and (ii) an graphical interface for efficient human supervision
(“Interact”). Evaluation involving recordings
of multiple moving mice shows that DIPLOMAT’s
automated method yields reductions in identity
swaps of 80 to 95% relative to leading methods,
and that these can then be almost entirely eliminated
with time-efficient human editing.
PDF