search

Kyfafyd / MirrorGAN

Reproduction of the paper MirrorGAN: Learning Text-to-image Generation by Redescription
MIT License
25 stars 4 forks source link

Issues with train #4

Closed GroundFinchOfGalapagos closed 3 years ago

GroundFinchOfGalapagos commented 3 years ago

I‘d like to train the model and when I try to run the GLAM main.py, the error occurs, I find it on the google,but i failed to solve the problem, can you teach me why this error occur and what can I do to solve it. Thanks!

image

Kyfafyd commented 3 years ago

I‘d like to train the model and when I try to run the GLAM main.py, the error occurs, I find it on the google,but i failed to solve the problem, can you teach me why this error occur and what can I do to solve it. Thanks!

image

are you running on bird dataset?

WhereISmyroad commented 3 years ago

I‘d like to train the model and when I try to run the GLAM main.py, the error occurs, I find it on the google,but i failed to solve the problem, can you teach me why this error occur and what can I do to solve it. Thanks! image

are you running on bird dataset?

Hello sir I train Image caption 700 times in my computer, but when I turn to train MirrorGAN get cap_loss : 423. How to distinguish whether the image caption model is correct? Thanks

Kyfafyd commented 3 years ago

I‘d like to train the model and when I try to run the GLAM main.py, the error occurs, I find it on the google,but i failed to solve the problem, can you teach me why this error occur and what can I do to solve it. Thanks! image

are you running on bird dataset?

Hello sir I train Image caption 700 times in my computer, but when I turn to train MirrorGAN get cap_loss : 423. How to distinguish whether the image caption model is correct? Thanks

Thanks for discussing! I would like to learn the Image Caption loss you get is?

WhereISmyroad commented 3 years ago

Thank you for your reply I use the code you provided. How do you modify it to make the generation model 5450x/ {I change the .json， but I think it is wrong to do so} Thanks

Kyfafyd commented 3 years ago

Thank you for your reply I use the code you provided. How do you modify it to make the generation model 5450x/ {I change the .json， but I think it is wrong to do so} Thanks

Sorry for reply so late... Can you describe your problem more clearly？

WhereISmyroad commented 3 years ago

Thank you for your reply I use the code you provided. How do you modify it to make the generation model 5450x/ {I change the .json， but I think it is wrong to do so} Thanks

Sorry for reply so late... Can you describe your problem more clearly？

Thanks for your reply The rnn_decoder.ckpt model you provided is 5453x300，how to change to 5450x300 when training STREAM Thank you very much.

Kyfafyd commented 3 years ago

Thank you for your reply I use the code you provided. How do you modify it to make the generation model 5450x/ {I change the .json， but I think it is wrong to do so} Thanks

Sorry for reply so late... Can you describe your problem more clearly？

Thanks for your reply The rnn_decoder.ckpt model you provided is 5453x300，how to change to 5450x300 when training STREAM Thank you very much.

I think this size mismatch is caused by using different dataset which has different vocabulary length. As for solution, you may pretrain STREAM model and STEM model at first. You can refer to this pretrain section in readme file.

GroundFinchOfGalapagos commented 3 years ago

thank for your reply, there was some problem with this mailbox so I haven't see these message until now I use the model provided by Attn-GAN,it's 5450x300 and we re-train a new model which can match the size very sorry for apply so late thank you very much for the discuss

At 2020-11-29 17:12:40, "Zhao Wang" notifications@github.com wrote:

Thank you for your reply I use the code you provided. How do you modify it to make the generation model 5450x/ {I change the .json， but I think it is wrong to do so} Thanks

Sorry for reply so late... Can you describe your problem more clearly？

Thanks for your reply The rnn_decoder.ckpt model you provided is 5453x300，how to change to 5450x300 when training STREAM Thank you very much.

I think this size mismatch is caused by using different dataset. As for solution, you may pretrain STREAM model and STEM model at first. You can refer to this pretrain section in readme file.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.