문제 제기: 기존까지 방법들 각 conv block에 정보를 전체적으로 무시하고 memory block을 도입하여 해결하려고 해도 이게 지속적으로 유지하기 힘듦. 그리고 다양한 featrures를 LR image로부터 뽑아내지 않음. 결국 연산량 증가뿐만 아니라 디테일도 잃음.

To address these drawbacks, we propose residual dense network (RDN) (Fig. 2) to fully make use of all the hierarchical features from the original LR image with our proposed residual dense block (Fig. 1(c)).

이러한 문제에 대응하기 위해 RDN을 제안함. LR 이미지로부터 모든 hierarchical features를 사용하고, Fig 1(c)에 나오는 residual dense block으로 구성됨.

We propose residual dense block (RDB) as the building module for RDN. RDB consists dense connected layers and local feature fusion (LFF) with local residual learning (LRL).

RDN을 구축하기 위한 RDB를 제안함. RDB는 dense connected layers와 local residual learning과 함께 local feature fusion으로 구성됨,

Concatenating the states of preceding RDB and all the preceding layers within the current RDB, LFF extracts local dense feature by adaptively preserving the information. Moreover, LFF allows very high growth rate by stabilizing the training of wider network. After extracting multi-level local dense features, we further conduct global feature fusion (GFF) to adaptively preserve the hierarchical features in a global way.

현재 RDB에 앞선 RDB의 states와 모든 앞선 layers를 합쳐, LFF는 정보 보존을 적용하여 local dense feature(Local feature fusion is a technique used in computer vision to combine local features extracted from an image into a single feature vector. This is done in order to improve the discriminative power of the features and to make them more robust to noise.)를 추출함. 추가적으로 LFF는 wider network의 안정적인 학습에 의해 매우 높은 비율로 성장을 허용함. 이후 multi-level local dense를 추출한 후global feature fusion을 적용함.

1. Residual Dense Network for Image SR

Network Structure

our RDN mainly consists four parts: shallow feature extraction net (SFENet), redidual dense blocks (RDBs), dense feature fusion (DFF), and finally the up-sampling net (UPNet).

RDN은 4가지 파트로 구성되었음. SFENet, RDBs, DFF, UPNet임.

SFE Net에 LR image를 넣음. 처음 입력은 단순 이것만 하지만 이후부터는 RDB에 입력을 함.

After extracting hierarchical features with a set of RDBs, we further conduct dense feature fusion (DFF), which includes global feature fusion (GFF) and global residual learning (GRL). DFF makes full use of features from all the preceding layers and can be represented as