search

knazeri / edge-connect

EdgeConnect: Structure Guided Image Inpainting using Edge Prediction, ICCV 2019 https://arxiv.org/abs/1901.00212
http://openaccess.thecvf.com/content_ICCVW_2019/html/AIM/Nazeri_EdgeConnect_Structure_Guided_Image_Inpainting_using_Edge_Prediction_ICCVW_2019_paper.html
Other
2.5k stars 530 forks source link

something wrong on testing #137

Open 1559145087 opened 4 years ago

1559145087 commented 4 years ago

I train the model on celeba-HQ in model 3, in the process of training, I found the sample is acceptable like below. 164000 However, when I test on the dataset, I found result all like this. celeba_01 there must something wrong on my test, hope to get your help, thanks

1559145087 commented 4 years ago

by the way, I found on dataloader, pytorch1.0.1 does't allow using sample while shuffle = True, so I change it to shuffle =False, and I tested on RTX2080Ti*2, cuda10 and I changed to pytorch1.0.0, I can set shuffle = True, but the test question is still on

1559145087 commented 4 years ago

@knazeri

1997Jessie commented 2 years ago

why my sample is only mask ,I put the train_gt.flist and train_structure.flist, I don't know what is problem in my config.yaml！SOS！

SouthCity commented 1 year ago

same problem.

TT-mouse commented 1 year ago

I train the model on celeba-HQ in model 3, in the process of training, I found the sample is acceptable like below. 164000 However, when I test on the dataset, I found result all like this. celeba_01 there must something wrong on my test, hope to get your help, thanks

I also encountered the same problem when testing. Have you solved it? I am looking forward to your reply.