Tao Hu

This is a intern project for Multiple object tracking.

Proposed a tracklet scoring model based on margin loss and rank loss, which provide a clear goal for quantifying tracklet quality. We, for the first time, demonstrate that contrastive learning can improve the tracklet consistency in data association.

Implemented a recurrent search-based optimization framework that remarkably exposes wrong associations to training. The training process follows a “searching-learning-ranking-pruning” pipeline. It tackles the problem of exposure bias existing in sequence modeling that is neglected by previous MOT works.

Implemented multiple hypotheses tracking (MHT) strategies to address the challenging problems of object occluded or missing. Validated on three benchmark datasets of MOT and achieved the state-of-the-art results.

This is a intern project for single object tracking.

Developed a feature-enhanced algorithm to deal with pervasive target occlusion, motion blur and fast motion in visual object tracking.

This proposed correlation filter based tracking method aggregates historical features in a spatial-aligned and scale-aware paradigm.

Evaluated this algorithm on benchmark datasets such as OTB and VOT, and achieved leading performance in real-time speed.

This is a project for multiple cell instance segmentation and tracking.

Proposed a video multi-cell instance segmentation algorithm based on mask propagation, which effectively utilizes the motion dependency of inter-frames;

Built motion model of Kalman filtering based on position and geometry features, which enables multi-cell tracking and mitotic detection.

Developed a GUI software for multi-cell segmentation, tracking and fluorescin measurement based on pyQt, which also supports image annotation.