search

htzheng / CM-GAN-Inpainting

CM-GAN for Image Inpainting
Apache License 2.0
241 stars 20 forks source link

About the network model and codes #1

Open song201216 opened 2 years ago

song201216 commented 2 years ago

Hello, your work is very amazing, but when will you release the network structure and training code? Thanks！

htzheng commented 2 years ago

Hi @song201216 , thank you for your interest in this project. We plan to release the online demo soon for inference. Since the model was developed during my internship, I am still at the stage to request permission for releasing the code.

alexmi256 commented 2 years ago

I hope you're successfull in your request to release. One thing I'm curious about is if you were able to test this with a faces dataset to see how it performs. Also, how long does training take given the hardware that you had?

htzheng commented 2 years ago

I hope you're successfull in your request to release. One thing I'm curious about is if you were able to test this with a faces dataset to see how it performs. Also, how long does training take given the hardware that you had?

Hi, thank you for your interest in our project! We will release the online demo of CM-GAN soon (probably in the next two months) and will most likely release the training code after releasing the online demo. Currently, we only test the model on Places2 but we are working on training an FFHQ model. Our model is trained on A100 GPUs for 1 week (similar to CoModGAN).

styler00dollar commented 2 years ago

How many GPUs were used?

htzheng commented 2 years ago

How many GPUs were used?

It uses 8 gpus.

roimulia2 commented 2 years ago

Any news about the pre-trained model? :) Loved the paper, would love to benchmark it.

htzheng commented 2 years ago

@roimulia2 Thank you for your interest in this work. I am gradually working towards releasing the model. Currently, the dataset, panoptic annotation, evaluation of CM-GAN, and the on-the-fly mask generation code are released. Will work on the online demo next.

htzheng commented 2 years ago

I hope you're successfull in your request to release. One thing I'm curious about is if you were able to test this with a faces dataset to see how it performs. Also, how long does training take given the hardware that you had?

Thank you for raising this and it is a very good point. We are also interested in training a model on FFHQ and will possibly work on it later.

surelon commented 2 years ago

Dear author, hello! I am also interested in the network model of your article, may I ask whether it will be updated on Github after the license is released? I do object detection, would like to quote this article, if convenient, can you send a copy to sdaujdxy2016ks@163.com, thank you very much!

raziehkaviani commented 2 years ago

Thank you for the amazing work and impressive results. I am wondering that when will you release the code? Thanks much.

shallweiwei commented 1 year ago

@htzheng Hello, could you release the training and inference code ?

htzheng commented 1 year ago

Since my company is developing products with CM-GAN, they are still deciding when to release the code. Sorry about this...

liangyuanyuan-lyy commented 1 year ago

Hello, I am very interested in what you have published. Can I publish the training code of CM-GAN?

Sanster commented 1 year ago

@htzheng I am very interested in the Lama-OT model. Is it possible to release the weight file of this model? Thank you.

lijain commented 1 year ago

看了你们的文章，感觉很有趣，不过有几点不是很清楚，可以方便解答下？ image 1 你们的Fout g也是会单独计算loss是？ 2 第二decoder是不引入encoder的信息，只拿取第一个decoder的Fine?