search

NUS-HPC-AI-Lab / Neural-Network-Parameter-Diffusion

We introduce a novel approach for parameter generation, named neural network parameter diffusion (p-diff), which employs a standard latent diffusion model to synthesize a new set of parameters
835 stars 43 forks source link

About Cond P-Diff #24

Open zhanglijun95 opened 2 months ago

zhanglijun95 commented 2 months ago

Hi Authors,

Thank you for your great work! It inspired me a lot! I'm really looking forward to your code for Cond P-Diff. May I know the estimated time for getting access to that?

Besides, I have a question about Cond P-Diff. I saw the CV task in this paper is style image generation and Cond P-Diff will generate parameters according to the conditions, namely the style image. I want to know when you test Cond P-Diff, do you give it the style image it is trained with, or a totally new/unseen style? For example, train the Cond P-Diff with 10 style-parameter pairs, and test with another 5 styles.

I noticed that in the Appendix, you mentioned the style-continuous dataset and the generalizability of Cond P-Diff to generate parameters for style in the range that is not in the trainset. But here I want to discuss with you that do you think it can generate parameters for a totally unseen style? Or do you have any insight about this?

Really appreciate your response and great work. Thank you!

Best, Lijun

breAchyz commented 2 months ago

Very good job! I also look forward to COND P-Diff!

Jinxiaolong1129 commented 2 months ago

Hi, thanks for your attention to our work. We will opensource by the end of Sep.

zhanglijun95 commented 2 months ago

Thank you for your confirmation! That's great.

zhanglijun95 commented 1 month ago

Hi, do we have any updates regarding the new code?

vishvak-ravi commented 3 weeks ago

Hi, thanks for your attention to our work. We will opensource by the end of Sep.

Any updates on the timeline for conditional P-diff?

Jinxiaolong1129 commented 3 weeks ago

Hi, due to GPU issues, we will update by next week.

zhanglijun95 commented 1 week ago

Hi, can we get access to the parameter autoencoder and the unet architecture used in conditional P-diff first?

Jinxiaolong1129 commented 1 week ago

Hi, I have sent all the code and dataset through email. Please check. We will reformat our code soon.

zhanglijun95 commented 1 week ago

I recieved it. Thank you so much!

srymaker commented 5 days ago

Hi, any updates on the timeline for conditional P-diff?

Jinxiaolong1129 commented 5 days ago

Hi, please email to jinxiaolong1129@gmail.com. I will share all the details.