From raw animal video to behavior understanding
This page organizes the research questions behind Junu's Lab: animal pose estimation, behavior recognition, video understanding, and annotation-efficient learning for animal behavior analysis.
Core research tracks
Annotation-efficient learning
activeHow can labeling cost be reduced when building animal behavior datasets?
Animal behavior datasets are expensive to label, so weak supervision, pseudo-labeling, semi-supervised learning, and active learning can make experiments more scalable.
Behavior recognition from animal video
activeHow can short and long animal behaviors be classified from video, pose sequences, or multimodal features?
Behavior recognition connects low-level visual signals to interpretable animal behavior analysis.
Animal pose estimation & keypoint trajectories
activeHow can animal body movement be represented from raw video using pose and keypoint trajectories?
Pose and trajectory representations reduce the complexity of raw video and provide a structured basis for behavior analysis.
From paper tracking to experiment logs
Research Radar is the weekly workflow that turns broad scans of papers, repositories, datasets, and benchmarks into selected reviews, experiment ideas, and project updates.
Current research status
Research tracks linked to projects and notes
Each track connects the research agenda to project artifacts, lab notes, paper reviews, and small experiments.
Annotation-efficient learning
active2 projects
Behavior recognition from animal video
active2 projects
Animal pose estimation & keypoint trajectories
active1 project