Joint detection and embedding (JDE) methods usually fuse the target motion information and appearance information as the data association matrix, which could fail when the target is briefly lost or blocked in multi-object tracking (MOT). In this paper, we aim to solve this problem by proposing a novel association matrix, the Embedding and GioU (EG) matrix, which combines the embedding cosine distance and GioU distance of objects. To improve the performance of data association, we develop a simple, effective, bottom-up fusion tracker for re-identity features, named SimpleTrack, and propose a new tracking strategy which can mitigate the loss of detection targets. To show the effectiveness of the proposed method, experiments are carried out using five different state-of-the-art JDE-based methods. The results show that by simply replacing the original association matrix with our EG matrix, we can achieve significant improvements in IDF1, HOTA and IDsw metrics, and increase the tracking speed of these methods by around 20%. In addition, our SimpleTrack has the best data association capability among the JDE-based methods, e.g., 61.6 HOTA and 76.3 IDF1, on the test set of MOT17 with 23 FPS running speed on a single GTX2080Ti GPU.
In this paper, we propose a deep residual sparse and cross-domain reconstruction network to reconstruct high-quality dynamic MR image sequences from undersampled data, which alternately learns to reconstruct signals in x-f domain and image domain in an iterative manner. Specifically, in the x-f domain, we decompose the unknown signal into the prediction signal and the residual signal. In order to sparsify the residual signal as much as possible, we exploit the powerful reasoning ability of neural networks to adaptively learn the residual signal and carry out the subsequent sparse coding processing. Furthermore, the spatiotemporal correlation of dynamic MR data in x-f space is used to recover the true signal from the aliasing signal. In the image domain, bidirectional ConvLSTM and dense connections have been incorporated in a U-net based reconstruction architecture to encourage the easier transfer of information. Gradually restoring sharp image details under different resolutions is realized in the decoding part of the U-net reconstruction architecture that is designed to avoid the over-smoothness issue. Experiments on vivo MR data confirm that the proposed method can achieve superior image reconstruction quality by preserving more fine details when compared to state-of-art methods.