iKUN: Speak to Trackers without Retraining
This work addresses the problem of efficiently integrating textual descriptions into multi-object tracking for researchers and practitioners, offering a plug-and-play solution that avoids retraining, though it is incremental as it builds on existing trackers.
The paper tackles referring multi-object tracking (RMOT) by proposing iKUN, a plug-and-play framework that enables communication with off-the-shelf trackers without retraining, achieving effectiveness verified on the Refer-KITTI dataset and introducing a new dataset, Refer-Dance, to advance the field.
Referring multi-object tracking (RMOT) aims to track multiple objects based on input textual descriptions. Previous works realize it by simply integrating an extra textual module into the multi-object tracker. However, they typically need to retrain the entire framework and have difficulties in optimization. In this work, we propose an insertable Knowledge Unification Network, termed iKUN, to enable communication with off-the-shelf trackers in a plug-and-play manner. Concretely, a knowledge unification module (KUM) is designed to adaptively extract visual features based on textual guidance. Meanwhile, to improve the localization accuracy, we present a neural version of Kalman filter (NKF) to dynamically adjust process noise and observation noise based on the current motion status. Moreover, to address the problem of open-set long-tail distribution of textual descriptions, a test-time similarity calibration method is proposed to refine the confidence score with pseudo frequency. Extensive experiments on Refer-KITTI dataset verify the effectiveness of our framework. Finally, to speed up the development of RMOT, we also contribute a more challenging dataset, Refer-Dance, by extending public DanceTrack dataset with motion and dressing descriptions. The codes and dataset are available at https://github.com/dyhBUPT/iKUN.