kpandey008/diffusevae

Official implementation of "DiffuseVAE: Efficient, Controllable and High-Fidelity Generation from Low-Dimensional Latents"

diffusevae's Issues

Pretrained Checkpoints

Hi,

Do you have the pretrained checkpoints available yet?

Regards,
Gaurav

Pipfile Incomplete

The pipfile doesnt contain any dependencies. If you install the virtualenv from it, it will not have the sufficient dependencies to run the training code.

noise conditional difussion training error

For noise conditional difussion VAE, I am getting the below error.

"wrapper.py", line 65, in training_step
    cond, x = batch
ValueError: too many values to unpack (expected 2)

FID score on cifar10 dataset

Hello, I trained your code, but after I trained 2850 epochs on cifar10, using form2 and not using GMM, the FID@50k I got was only 4.00, which is much different from the 2.86 in your paper. I'm wondering why this is happening? Can you give your complete list of hyperparameters during training? Thank you！

FID evaluation details

Hi, I have some questions about FID evaluations for CIFAR-10 and CelebA-64.
May I know how many images you generate to compute fid for CIFAR-10 and CelebA-64, respectively?
And which split of the CIFAR-10 and CelebA-64 datasets do you use to compute fid?
I am assuming that the training set of CIFAR-10 (50,000 images), and the whole dataset of CelebA (202,599 images) are used to compute fid. Is that correct?

Image denoising

Hi,

I would like to reproduce the results for image denoising as depicted in Figure 9 (Right image).

May I know how to format the dataset to apply your code?

Many thanks,
Vinod

CelebHQ sampling

Hi,

I would like to make some samples based on CelebHQ dataset when inferencing. When runing the scripts provided, I got errors saying that the model size mismatched

May I know how to use the CelebHQ checkpoints，how to set parameters in the script?

Thanks a lot.

What is the method of selecting the model architectures?

I want to reproduce it on MNIST and ImageNet, but I don't know how to choose the network architecture of the VAE. Do you have any suggestions?

data name

Hi,

Why the argument data.name='recons' for celeba dataset in README?

I can find seperate file for each dataset (celeba.py) are available already. what is the difference between celeba.py and recons.py and when to use it?

Thanks,
Vinod

sample a image directly from a low dimension to a high dimension

hi thank u for your wonderful work !

i have already try my own the model on my own dataset ,and i got the checkpoints

i am trying to sample a image directly from a low dimension ,like 1024，to 128*128 dimension image using Diffuse VAE
but i found that there is no corresponding code for this task in sample .py or sample_cond
is this feasible ?
How should I implement it？
should i change the code ?

i really hope i could get UR help
thank you so much !

Controllable synthesis

Thank you for the great repository!

Could you please show us the implementation code when you add or subtract the meaningful concepts, as in figure.14, 15 of the arxiv version?
Also, is the checkpoint of the model in that case the same as the published one?

Thank you in advance.