search

cripac-sjx / SEA-T2F

Multi-caption Text-to-Face Synthesis: Database and Algorithm
30 stars 4 forks source link

pretrain model path may be not correct? #1

Open chaowentao opened 2 years ago

chaowentao commented 2 years ago

Hi, your work is excellent. I am interested in it and want to generate the result. While I find the pretrain models' links are the same as DAMSM models link. so it may be some wrong. please correct it. thank you. 微信截图_20220109153927

RachelWang122 commented 1 year ago

Hello, It's a very valuable job，Can you provide a link to the pre-trained model for netG，thank you. @cripac-sjx

chaunceywx commented 1 year ago

Hi, your work is excellent. I am interested in it and want to generate the result. While I find the pretrain models' links are the same as DAMSM models link. so it may be some wrong. please correct it. thank you. 微信截图_20220109153927

Hello, Have you obtained the correct download address for the pretrain model?

chaowentao commented 1 year ago

@chaunceywx Oh, I don't obtain the correct download address. I'm waitting for the author's reply.

chaowentao commented 1 year ago

@chaunceywx Oh, I don't obtain the correct download address. I'm waiting for the author's reply.

Excuse me, have you tried training your own netG? Does this code really work? I'm not looking forward to the author's response anymore and trying to train my own netG. But I found some bugs. e.g. wd_embs, s_emb = text_encoder(captions[i], sort_cap_lens, hidden) in trainer.py, line 221. The hidden is on the cpu, but the captions[i] is on the gpu. This makes the code not work.

Sorry, I don't try train my own model because it's not my research direction.