Liu D, Wen B, Fan Y, et al. Non-local recurrent network for image restoration[C]//Advances in Neural Information Processing Systems. 2018: 1673-1682.
1. Overview
1.1. Motivation
- existing methods do not explicitly use self-similarity properties in images
In this paper, it proposed non-local recurrent network (NLRN).
2. Methods
2.1. Non-local Module
2.2. Non-Local Recurrent Network
- x. input state
- y. output state
s. recurrent state
s^0. function of input Image I.
- x^t = 0, f_input(x) = 0
- output state y^t calculate only at time T
- s^0. add an identity path from the very first state
- s_corr. add a residual path of deep feature correlation between each location and its neighbourhood from previous state