SRNTT Project Realization

This is a Tensorflow implemention of SRNTT. (Zhifei Zhang, Zhaowen Wang, Zhe Lin, and Hairong Qi, "Reference-Conditioned Super-Resolution by Neural Texture Transfer", arXiv:1804.03360v1, 2018.) FOR STUDY ONLY. ALL THE RIGHTS BELONG TO THE ORIGINAL AUTHORS.

OFFLINE BRANCH OF SRNTT

Module 1: Texture Extractor (''test_vgg19.py'')

Description: pre-trained VGG19 model.

Function: vgg19_pretrained()

Input: image, 160 x 160 x 3

Output: feature matrix of conv3_1 (40x40x256)

feature matrix of conv5_1 (10x10x512)

Module 2: Patch Matching & Texture Swapping (''patch_match.py'')

Description: swap low-res image with high-res patches.

Function: Fun_patchMatching()

Input: M_LR, M_LRef, M_Ref (40x40x256)

Output: M_t, M_s (40x40x256)