20xx行人再識別年度進(jìn)展回顧-valse(文件)
【正文】 2022 Outline ? Train/test on the same domain ? Unsupervised domain adaptation ? Person search ? Other problems ? Future research questions Person reid originates from person tracking。 After years of study, person reid features are improving tracking accuracies. Reid features are applied to multitarget multicamera tracking Triplet loss, % IDF1 (Ristani et al., CVPR 2022) Partbased model, % IDF1 (Yoon et al., TIP 2022) Random forest, % IDF1 (Jiang et al., ACM MM 2022) Result on the DukeMTMC dataset Siamese work, % MOTA （ Ma et al., ACCV 2022） StackNetPose, % MOTA (Tang et al., CVPR 2022) TrackletNet, % MOTA (Wang et al., arXiv 2022) Triplet loss, % MOTA (Shen et al., arXiv 2022) Reid feature learning is applied to Multiple object tracking (MOT) Open question: will better reid features lead to better tracking accuracy? Learning from synthetic data for person reid SOMAset and SOMA (Barbosa et al., CVIU 2022) changing clothing and pose Synthetic training data can help to initialize deep works different body shapes SyRI (Bak et al., ECCV 2022) The diversity of synthetic data can help improve the generalization performance of reID models PersonX (Sun et al., CVPR 2022) ? How do viewpoint distributions in the Training Set affect model learning? ? How does Query viewpoint affect retrie
環(huán)評公示相關(guān)推薦
鄂ICP備17016276號-1