alihan/sketch-to-photograph-with-GANs
1
0
Fork 0
mirror of https://github.com/inzva/sketch-to-photograph-with-GANs.git synced 2021-06-01 09:32:15 +03:00
Code Issues Projects Releases Wiki Activity

change img paths in md

Browse Source
This commit is contained in:
cemre
2019-11-09 23:35:09 +03:00
parent f469d172d8
commit 46c5c4a154
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
README.md
View File

@@ -36,12 +36,12 @@ Run **binarization.py** to get final edge maps.
### Baseline Model
The model as seen in the paper by Chen and Hals.
![Fig 1.1 (source: SketchyGAN by Chen and Hals)](model.png)
![Fig 1.1 (source: SketchyGAN by Chen and Hals)](misc/model.png)
It follows a simple encoder-decoder structure for the generator and encoder structure for the discriminator. We used a one layer discriminator since a multi-layered approach proved rather complex for the problem we are dealing with.
Uses MRU layers instead of convolutional layers, described by the image below:
![Fig 1.2 (source: SketchyGAN by Chen and Hals)](mru.png)
![Fig 1.2 (source: SketchyGAN by Chen and Hals)](misc/mru.png)
MRU provides continuous learning on the input image by taking in the input image with different sizes at every layer.
Skip connections also contribute to the preservation of information by concatenating the output of encoder layers to the decoder inputs at every layer (as seen by the dashed lines in Fig 1.1).