vmTracking enables accurate identification in crowded environments (IMAGE)
Caption
Conventional markerless tracking methods struggle with body part misestimations or missing estimates in crowded spaces. In vmTracking, markerless multi-animal tracking is performed on a video containing multiple individuals. The resulting tracking output may not always be fully accurate. However, since some of these markers are extracted and used as virtual markers for individual identification, high overall accuracy at this stage is not required. By applying single-animal DeepLabCut to the generated virtual marker video, more accurate pose-tracking results can be obtained compared to conventional methods.
Credit
Hirotsugu Azechi from Doshisha University, Japan
Usage Restrictions
Cannot be used without permission
License
Original content