avalonstrel/GatedConvolution_pytorch

About pretrained model

huangqianfirst opened this issue · 14 comments

Could you please update the complete pretrained model. I've test the corase model.The result is not good.
44
39
Thank you very much.

UPML commented

+1
I have the same results with pretrained model.
And also what about dark areas at generated images?
40
44
39

Thanks, the model is trained before the model structure changed, so I will retrain a model using the current script.

UPML commented

Ok, thanks. How much GPU time do you spent on training?

@avalonstrel I use the TensorFlow version and alsoencountered this problem.Have you solve this problem,and what do you think causes the result?

why i download the pretrained models, but it is broken?? and i can not open it ?

I have trained 'Free_from Image Inpainting with Gated Convolution' from scratch ,here is my result
https://github.com/huangqianfirst/Free_from-Image-Inpainting-with-Gated-Convolution

@huangqianfirst how did you create mask file while training?

janzd commented

@huangqianfirst Did you use the implementation in this repository? If yes, what settings did you use to be able to produce such good results as you show?

@kurapan .No ,i implemente by myself based DeepFillv1 (https://github.com/JiahuiYu/generative_inpainting).

@sporterman .You can get the advice about how to create mask from here:JiahuiYu/generative_inpainting#62.

@huangqianfirst Is that a valid set result? and also do have any idea to open the implementation?

@DongHwanJang . 1. It's test results. 2. Not yet.

@huangqianfirst how long did it take to train?

@DongHwanJang . About five days using Titan X, one GPU.