search

liuzhengzhe / DreamStone-ISS

This is the project for DreamStone: TPAMI & ISS: ICLR 2023 spotlight
https://liuzhengzhe.github.io/DreamStone.github.io/
43 stars 3 forks source link

Some questions about your paper #3

Closed WayneCV closed 1 year ago

WayneCV commented 1 year ago

Greets, I read your paper and your code, it is a good job. But I have some questions:

  1. For each training text, did you retrain the model for the corresponding text separately? For example, in Table 5 from your paper, whether you train one model with the text "A green SUV" to show the result with "A green SUV"?

  2. Suppose we have trained one model with the training text "A green SUV", if we want the result with "A wooden boat", do we need to retrain a model with the text "A wooden boat" to get the result on "A wooden boat"?

Thanks in advance.

liuzhengzhe commented 1 year ago

Thanks for your message.

  1. All the cases share the same stage-1 model, but each text has its own stage-2 model. Note that the training of stage-2 is fast (around 85 seconds).
  2. We need to run stage 2 training for "a wooden boat". The training should still be initialized from the stage 1 model. For the case "a wooden boat", we may need a bit more iterations (--iter 30).

Zhengzhe

WayneCV @.***> 于2023年4月20日周四 10:36写道：

Greets, I read your paper and your code, it is a good job. But I have some questions:

1.

For each training text, did you retrain the model for the corresponding text separately? For example, in Table 5 from your paper, whether you train one model with the text "A green SUV" to show the result with "A green SUV"? 2.

Suppose we have trained one model with the training text "A green SUV", if we want the result with "A wooden boat", do we need to retrain a model with the text "A wooden boat" to get the result on "A wooden boat"?

Thanks in advance.

— Reply to this email directly, view it on GitHub https://github.com/liuzhengzhe/ISS-Image-as-Stepping-Stone-for-Text-Guided-3D-Shape-Generation/issues/3, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB7WPDEOSDGI2IYHWQ3OMJDXCCOJJANCNFSM6AAAAAAXE3PDAY . You are receiving this because you are subscribed to this thread.Message ID: <liuzhengzhe/ISS-Image-as-Stepping-Stone-for-Text-Guided-3D-Shape-Generation/issues/3 @github.com>

WayneCV commented 1 year ago

Thanks for your message. 1. All the cases share the same stage-1 model, but each text has its own stage-2 model. Note that the training of stage-2 is fast (around 85 seconds). 2. We need to run stage 2 training for "a wooden boat". The training should still be initialized from the stage 1 model. For the case "a wooden boat", we may need a bit more iterations (--iter 30). Zhengzhe WayneCV @.***> 于2023年4月20日周四 10:36写道： … Greets, I read your paper and your code, it is a good job. But I have some questions: 1. For each training text, did you retrain the model for the corresponding text separately? For example, in Table 5 from your paper, whether you train one model with the text "A green SUV" to show the result with "A green SUV"? 2. Suppose we have trained one model with the training text "A green SUV", if we want the result with "A wooden boat", do we need to retrain a model with the text "A wooden boat" to get the result on "A wooden boat"? Thanks in advance. — Reply to this email directly, view it on GitHub <#3>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB7WPDEOSDGI2IYHWQ3OMJDXCCOJJANCNFSM6AAAAAAXE3PDAY . You are receiving this because you are subscribed to this thread.Message ID: <liuzhengzhe/ISS-Image-as-Stepping-Stone-for-Text-Guided-3D-Shape-Generation/issues/3 @github.com>

Thanks!