Texture Synthesis comparison of methods and results of the perceptual test

Description :

The zip folder contains the 20 reference images of resolution 1024*1024 with the synthesis obtained by different methods.
Those 20 references images (in the folder named References) are the one used for the perceptual test.
The synthesis are in the folder corresponding to the name of the reference images.

The DisplacementMaps folder contains the displacement maps for the different methods we compared in the main file.

Nomemclature :
Each synthesis is followed by a postfix that refers to a given methods as follows :
- DCor : Deep Corr Sendick et al. 2017
- EfrosFreeman : Efros et al. 2001
- EfrosLeung : Efros et al. 1999
- autocorr : Autocorrelation [Our]
- MSInit : Our Multi resolution strategy [Our]
- Gatys : Gatys et al. 2015
- spectrumTFabs_eps10m16 : use of a spectrum constraints (Liu et al. 2016)
- Ulyanov : ulyanov et al. 2016
- _Snelgorove_MultiScale_o5_l3_8_psame : Snelgrove 2017

Our method :

Our methods are defined in the following research paper : "High resolution neural texture synthesis with long range constraints Gonthier et al. 2020" https://arxiv.org/abs/2008.01808

We introduce a simple multi-resolution framework that efficiently accounts for long-range dependency. Then, we show that additional statistical constraints further improve the reproduction of textures with strong regularity. This can be achieved by constraining both the Gram matrices of a neural network and the power spectrum of the image. Alternatively one may constrain only the autocorrelation of the features of the network and drop the Gram matrices constraints.

Perceptual Test :

The three csv files contains the results of the perceptual test conducted during our study to compared the differents methods.

The files are named Number_of_wins_both.csv, Number_of_wins_local.csv and Number_of_wins_global.csv.