Error-Aware Density Isomorphism Reconstruction for Unsupervised Cross-Domain Crowd Counting

This paper focuses on the unsupervised domain adaptation problem for video-based crowd counting, in which we use labeled data as source and unlabelled video target domain. It is challenging there a huge gap between no annotations of samples are available The key issue how to utilize videos knowledge learning transferring from To tackle this problem, propose novel Error-aware Density Isomorphism REConstruction Network (EDIREC-Net) cross-domain counting. EDIREC-Net jointly transfers pre-trained counting model domains using density isomorphism reconstruction objective models erroneousness by error reasoning. Specifically, flows consecutive, maps adjacent frames turn out be isomorphic. On basis, regard self-supervised signal transfer different domains. Moreover, leverage an estimation-reconstruction consistency monitor suppress unreliable reconstructions during training. Experimental results four benchmark datasets demonstrate superiority proposed method ablation studies investigate efficiency robustness. code at https://github.com/GehenHe/EDIREC-Net.