sherrylone / pqdiff Goto Github PK

MUST WE TRAIN OWN MODEL? WILL YOU RELEASE YOUR MODEL WEIGHTS?THANKS~

Hi, I downloaded the google drive data from the repository you mentioned. How to split this data into train and test so as to reproduce your results. I am reluctant to download Baidu softwares to download the split data.

TIA

Hello, according to the UViT network in the code, if the image size during training and output can only be the shape of (192，192)? But in the article, I remembered that we can denote a (w, h) as the predefined resolution to be generated. Can this be achieved without adjusting the network structure? Thank you so much for your reply.

Hello, i am getting problem in running the model as it asks for the missing file at the path "assets/stable-diffusion/autoencoder_kl.pth".

Will the model weight be released?

Hi authors,

Great work!

I did not modify your code and just played with the scenery dataset. The training stages are all fine.
However, when I moved to the sampling stage to see the outputs, it gives me results like this:

Seems like I am not able to get {Outpainting in an arbitrary position}

Could you please comment on that?

Many thanks,
Olivia

i am just letting you know,

https://github.com/Sherrylone/PQDiff/blob/e9ce669fe34f2ae43453a0f65d2ee6cd836febba/dataset/pos.py#L66C3-L66C54

here dtype is set to np.float i believe it should be np.float32

Very interested in this interesting work，It is hoped that the model can be provided with weights for testing.

I trained the model with the scenery dataset and when i am trying to generate an extended image it is creating a colorful noisy image. Not able to figure out what actually the issue is. I think there is something wrong in the decoder part. Please find the attached files for the reference.

请问在采样阶段这个评价文件在哪里下载