2.1, pay attention to the network based on parallel context scenario text image super-resolution

Scene text image super-resolution based on parallel contextual attention network

Scene Text Image Super-Resolution via Parallelly Contextual Attention Network

ACMMM 2021

https://github.com/Vill-Lab/PCAN

SR, image super resolution super resolution

Data
The main work of this paper is to construct a new dataset: the real scene text SR dataset, called TextZoom.It contains a pair of real low-res and high-res images

Model
Modification of the SRResNet model
Parallel contextual attention network, mainly changing the basic unit of RNN, parallel contextual attention block (PCAB)
Intuitively, the horizontal direction modelingUsed to build character-to-character
dependencies, while vertical modeling is used for texture context within characters.