Deep Learning-Based Identity Preserving Labeled-Object Multi-Animal Tracking
Deep Learning-Based Identity Preserving Labeled-Object Multi-Animal Tracking
DIPLOMAT provides a multi-animal tracking/pose-estimation interface that (1) reduces identity swaps and body part losses, and (2) simplifies correction of automated tracking errors. After pre-processing a video with a pose estimation tool (currently supporting SLEAP and DeepLabCut packages), DIPLOMAT computes a multi-animal maximum-probability trace to Track multiple animals (and their body parts); it then provides an intuitive and memory/time efficient Interact interface to edit and re-track as needed. DIPLOMAT differs from other multi-animal tracking packages by working directly off of confidence maps instead of running peak detection, allowing for more accurate tracking results.
Here we can make precise adjustments to our model's predicted points and re-run tracing.
Learn more about DIPLOMAT's functionalities through our video guides.
Learn how to install and configure DIPLOMAT using either conda or pip.
Introduction on how to track videos with DIPLOMAT, leveraging models created through SLEAP or DeepLabCut (DLC).
Overview of the interactive tracking interface in DIPLOMAT. Demonstrates how to adjust visual settings, predicted tracks, and re-run tracing.
Demonstrates DIPLOMAT's tweak interface. The tweak interface allows for quickly touching up finished tracks, removing swaps and making minor point adjustments.