Skip to content

Updated image sources #84

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
francotheengineer wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Updated image sources #84

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions TUTORIAL.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@ Unsupervised image-to-image translation concerns learning an image translation m

### Algorithm

<img src="https://raw.githubusercontent.com/NVIDIA/UNIT/master/docs/shared-latent-space.png" width="800" title="Assumption">
<img src="https://raw.githubusercontent.com/mingyuliutw/UNIT/master/docs/shared-latent-space.png" width="800" title="Assumption">

The unsupervised image-to-image translation problem is an ill-posed problem. It basically aims at discovering the joint distribution from samples of marginal distributions. From the coupling theory in probability, we know there exists infinitely many possible joint distributions that can arrive to two given marginal distributions. To find the target solution, one would have to incorporate the right inductive bias. One has to use additional assumptions. UNIT is based on the shared-latent space assumption as illustrated in the figure above. Basically, it assumes that latent representations of a pair of corresponding images in two different domains share the same latent code. Although we do not have any pairs of corresponding images during training, we assume their existences and utilize network capacity constraint to encourage discovering the true joint distribution.

Expand Down Expand Up @@ -89,7 +89,7 @@ The results are stored in `results/gta2city` folder. You should see images like

| Input Photo | Output Photo |
|-------------|--------------|
| <img src="https://raw.githubusercontent.com/NVIDIA/UNIT/master/results/gta2city/input.jpg" width="384" title="Input"> | <img src="https://raw.githubusercontent.com/NVIDIA/UNIT/master/results/gta2city/output.jpg" width="384" title="Output"> |
| <img src="https://raw.githubusercontent.com/mingyuliutw/UNIT/master/docs/input.jpg" width="384" title="Input"> | <img src="https://raw.githubusercontent.com/mingyuliutw/UNIT/master/docs/output.jpg" width="384" title="Output"> |

Run the following command to translate Cityscape images to GTA5 images

Expand All @@ -99,6 +99,6 @@ Run the following command to translate Cityscape images to GTA5 images

| Input Photo | Output Photo |
|-------------|--------------|
| <img src="https://raw.githubusercontent.com/NVIDIA/UNIT/master/results/city2gta/input.jpg" width="384" title="Input"> | <img src="https://raw.githubusercontent.com/NVIDIA/UNIT/master/results/city2gta/output.jpg" width="384" title="Output"> |
| <img src="https://raw.githubusercontent.com/mingyuliutw/UNIT/master/docs/input.jpg" width="384" title="Input"> | <img src="https://raw.githubusercontent.com/mingyuliutw/UNIT/master/docs/output.jpg" width="384" title="Output"> |