Closed Kyfafyd closed 4 years ago
Hi, it is pre-trained on AttnGAN. You can find their repository here https://github.com/taoxugit/AttnGAN.
Thanks very much for reply!
I know that text_encoder.pth and image_encoder.pth are pretrained on AttnGAN. But I wonder how to train decoder-1-700.ckpt and encoder-1-700.ckpt?
Thanks very much!
-----原始邮件----- 发件人:Haoliang notifications@github.com 发送时间:2020-03-14 22:45:33 (星期六) 收件人: Haoliang-rp/MirrorGAN_Reproduction MirrorGAN_Reproduction@noreply.github.com 抄送: Kyfafyd Kyfafyd@zju.edu.cn, Author author@noreply.github.com 主题: Re: [Haoliang-rp/MirrorGAN_Reproduction] Can you show me how to train decoder-1-700.ckpt and encoder-1-700.ckpt? (#2)
Hi, it is pre-trained on AttnGAN. You can find their repository here https://github.com/taoxugit/AttnGAN.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.
Hi, the decoder-1-700 and encoder-1-700 are actually trained on https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/03-advanced/image_captioning.
Thanks very much.
But I wonder how you preprocess data for training, especially json file(captions). I have been confused for some time.
-----原始邮件----- 发件人:Haoliang notifications@github.com 发送时间:2020-03-14 23:31:37 (星期六) 收件人: Haoliang-rp/MirrorGAN_Reproduction MirrorGAN_Reproduction@noreply.github.com 抄送: Kyfafyd Kyfafyd@zju.edu.cn, Author author@noreply.github.com 主题: Re: [Haoliang-rp/MirrorGAN_Reproduction] Can you show me how to train decoder-1-700.ckpt and encoder-1-700.ckpt? (#2)
Hi, the decoder-1-700 and encoder-1-700 are actually trained on https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/03-advanced/image_captioning.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.
Hi, the preprocessed bird dataset provided in the original repo of MirroGAN could be used as the reference. It should be compatible with the data loader in the image captioning model.
Thanks, I actually know that.
But I have been trying to preprocess a json file(refer to coco dataset). After that, the problem went into an error like this.
-----原始邮件----- 发件人:Haoliang notifications@github.com 发送时间:2020-03-15 00:04:12 (星期日) 收件人: Haoliang-rp/MirrorGAN_Reproduction MirrorGAN_Reproduction@noreply.github.com 抄送: Kyfafyd Kyfafyd@zju.edu.cn, Author author@noreply.github.com 主题: Re: [Haoliang-rp/MirrorGAN_Reproduction] Can you show me how to train decoder-1-700.ckpt and encoder-1-700.ckpt? (#2)
Hi, the preprocessed bird dataset provided in the original repo of MirroGAN could be used as the reference. It should be compatible with the data loader in the image captioning model.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.
I have found problem, thanks!
Thanks very much!