OCTRON - a general purpose segmentation and tracking pipeline for behavioral experiments

Quantifying animal behavior across species with complex, deformable body plans requires capturing full-body shape rather than sparse skeletal keypoints. Yet accessible, end-to-end tools that take researchers from mask annotation through model training to reliable multi-object tracking are lacking. Here we present OCTRON, a fully trainable pipeline for multi-object segmentation, detection, and tracking that is broadly applicable across experimental setups and organisms. OCTRON leverages large segmentation foundation models to rapidly generate experimenter-supervised annotations from minimal input, overcoming the manual labeling bottleneck of supervised segmentation. These annotations train compact, task-specific models: segmentation networks for extracting pixel-level information such as shape and appearance, or lightweight detection networks for fast centroid tracking. Coupled with multi-object trackers and an interactive track-curation interface in napari, OCTRON enables complete behavioral quantification with minimal manual intervention. We validate OCTRON across species from transparent marine annelids to camouflaging cuttlefish, and in multi-species behavioral assays.