train and paper details

[moon-bright-zhen]

Hello, thank you for your great work!
I'm very interested in this excellent work. I have the following two questions:

1. Step 3 of the training scene Craft model part executes
   ns-train ${method_name}.
   Does that mean I need to execute ns-train method.py ?

2. It is mentioned in the paper that the input is text and BBS. Does the BBS include camera trajectory? When users use BBS, what form should be input and what content should be included?

Looking forward to your examples code next. I will continue to pay attention to your excellent work.Thank you!

[OrangeSodahub]

Hi, thanks for your interest.

1. method_name is specified in method.py:

   SceneCraft/scenecraft/configs/method.py

   Line 113 in b15c078

   method_name="nerfacto-huge2",

2. Yes you should include the camera trajectories. Please refer to Step1 to check the input format, the camera trajectory and descriptions of 3d layout are written in .json files. You can generate them from the dataset, or drawn by yourself using our webgui.

Please let me know if you have further questions.

[moon-bright-zhen]

thanks for your reply! Want to know when you can release layout examples' code and detailed instructions for generation and visualization?

[OrangeSodahub]

Will be soon! I need to manage them well later. BTW, layout examples are just some .json files input (from dataset or create by myself). All codes are already available now.

[SCUTykLin]

thanks for your reply! Want to know when you can release layout examples' code and detailed instructions for generation and visualization?

hello, do u try their model on new layout? In my testing, I've noticed that the model can only generate in situations where there is complete absence of occlusion. If u work, could u share your layout example? Thanks