Skip to content
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.

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

Dataset Arrangement #2

Open
akashdexati opened this issue Apr 17, 2018 · 6 comments
Open

Dataset Arrangement #2

akashdexati opened this issue Apr 17, 2018 · 6 comments

Comments

@akashdexati
Copy link

Hi NVlabs
Can you please explain the dataset arrangement.
1). The folders name TrainA - TrainB - TestA - TestB. What type of images are we supposed to place.
2). How do we specify the style images. ?

@xunhuang1995
Copy link
Contributor

  1. You can place the following types of images.
    IMG_EXTENSIONS = [
    '.jpg', '.JPG', '.jpeg', '.JPEG',
    '.png', '.PNG', '.ppm', '.PPM', '.bmp', '.BMP',
    ]
  2. You can specify style images at test time as described in the "Example-guided Translation" section in the user manual page. At training time you don't need to specify style images.

@akashdexati
Copy link
Author

TrainA & TrainB are two different folders. Is it related to source(target) to target(source).
Can you please provide any sample image that is to be placed in these folders. ?

@mingyuliutw mingyuliutw reopened this Apr 19, 2018
@mingyuliutw
Copy link
Collaborator

@akashdexati Please check datasets folder.
We now provide two interfaces: one is folder-based and the other is list based. Check out edges2handbags_folder.yaml and edges2handbags_list.yaml for usage examples.

@AidasK
Copy link

AidasK commented Sep 20, 2018

Datasets structure could be explained in the usage.md because now it's a bit confusing.

As I can see in datasets/demo_edges2handbags:
trainA and trainB are random unrelated source and target images. (Unpaired data as stated in paper)
testA and testB are related source and target images. (Paired data)

Can testA and testB be unpaired too?

@abhishekcvedia
Copy link

abhishekcvedia commented Dec 26, 2018

I have the same doubt as @AidasK
Are we supposed to have paired dataset in trainA and trainB also in testA and testB?
@akashdexati @mingyuliutw @xunhuang1995

@Cuky88
Copy link

Cuky88 commented Dec 26, 2018

Let's consider you have 2 sets of image collections (unpaired) of a national park. One collection has images of the park in summer (domain 1) and the other one images in winter (domain 2). Than, you can put domain 1 in trainA and domain 2 in trainB. For testing you will also need images, therefore you take some of your domain 1 images from trainA and put them in testA, and you do the same for domain 2, take some images from trainB and put them in testB. The images in trainA and trainB are unpaired, so are the images in testA and testB.

It is just important that the 2 image collections (summer and winter) show the same content, e.g. the national park, so they don't need to be paired.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

6 participants