Multi-local Collaborative AutoEncoder

Twinned Residual Auto-Encoder (TRAE)—A new DL architecture for denoising super-resolution and task-aware feature learning from COVID-19 CT images

“…In Choi, Jeong, Lee, and Lee (2021) , an architecture is proposed, which relies on the a priori assumption of self-similar LR/HR image pairs and then exploits a set of Local Collaborative AEs (LOCA) for learning self-similar image patches. Along the same research line, in Chu, Wang, Liu, Gong, and Li (2022) , the Multi-Local Collaborative AE (MC-AE) is proposed for SISR. Similar to LOCA, also the MC-AE architecture is layer-wise trained, with higher-order AEs that are embedded into the lower-order ones.…”

“…Similar to LOCA, also the MC-AE architecture is layer-wise trained, with higher-order AEs that are embedded into the lower-order ones. Overall, as the proposed TRAE, the AE-based architectures in Choi et al (2021) and Chu et al (2022) are based on multiple AEs that work in a cooperative way. However, unlike TRAE, these architectures rely on the a priori assumption of local/non-local inter-patch self-similarity, which may fall short in medical imaging ( Greenspan, 2009 ).…”