DIPLOMAT: multi-animal tracking with efficient manual editing
- Open access
DIPLOMAT reduces animal identity swaps by over 75% in multi-mouse videos through advanced deep learning and efficient manual editing.
- Why it matters: Accurate, automated tracking of multiple animals remains challenging due to occlusion and identity loss, limiting behavioral analysis and requiring extensive manual correction.
- What they did: The team developed DIPLOMAT, a deep learning-based tracker built on DeepLabCut and SLEAP, which splits probability fields and incorporates cross-frame movement for improved identity preservation, with a user-friendly interface for minimal manual editing.
- The result: DIPLOMAT significantly enhances multi-animal tracking accuracy, enabling more reliable behavioral studies and reducing manual correction efforts, with open-source code available for broader use.