xingzhehe/ganseg

Code for CVPR 2022 paper GANSeg: Learning to Segment by Unsupervised Hierarchical Image Generation

ganseg's Issues

missing files for celeba preprocessing

Hi,

We are highly interested by your work, yet we can't find the following files on the download links you provided for celeba :

data/celeba_wild_raw/MAFL/celebA_training.txt
data/celeba_wild_raw/MAFL/training.txt
data/celeba_wild_raw/MAFL/testing.txt

Since these file seem mandatory to run the celeba dataset preprocessing, can you please indicate where we can download them.

Many thanks in advance,
Romain

stragen generation with provided checkpoints

Hi, thanks for sharing the good work. I used the provided checkpoints celeb_wild_k8 and taichi_k10 to generate images. It seems both the results are strange and are not natural images. I show some cases in the following.

celeb

taichi

BTW, I use the torch version 1.6.

missing example of training a gan "without requiring supervision of masks or points"

Hi,

We were able to run your gan on the celeba example but we are struggling to apply it on raw images (without any supervision of mask or keypoint).

After careful reading of your works, we understand the main interest of your model is to generate keypoints then masks then segmented images from training on a dataset with only raw images, without any mask or keypoints in the training set. Can you please confirm this and provide us with a version of your code that can be applied on a folder containing only raw 256x256 images ?

Many thanks again for your work and reactivity,
Romain

Important files of the datasets are missing

When I preprocess the datasets of Celeba, cub, and flowers, some of the important files are missing, such as ''list_imsize_celeba.txt'', ''MAFL/celebA_training.txt'', ''train_val_test_split.txt'', etc. Can you provide all these files of datasets?