search

maturk / dn-splatter

DN-Splatter: Depth and Normal Priors for Gaussian Splatting and Meshing
https://maturk.github.io/dn-splatter/
Apache License 2.0
492 stars 32 forks source link

background color #50

Open yungommi opened 5 months ago

yungommi commented 5 months ago

I scanned an object at the center and applied background removal, making all the backgrounds white. However, when I tested your code, white points from the background were also reconstructed. How can I fix this issue? Should I use a different background color when removing the background in my images to avoid this problem? i don't want background to be reconstructed. only object at the center is needed

maturk commented 5 months ago

Hi, currently the background is set to random by default. Can you try adding the flag --pipeline.model.background-color white when training with ns-train dn-splatter [options] --pipeline.model.background-color white --data path_to_data/ this will set the background color to white.

One other option if this does not work is to use masks while trianing.

Let me know if any of these help. You can also post images/screenshots so I can see the results better. I have not tried training with different backgrounds yet.

ruiqiyan commented 2 weeks ago

Hi, currently the background is set to random by default. Can you try adding the flag--pipeline.model.background-color whitewhen training withns-train dn-splatter [options] --pipeline.model.background-color white --data path_to_data/this will set the background color to white.

One other option if this does not work is to use masks while trianing.

Let me know if any of these help. You can also post images/screenshots so I can see the results better. I have not tried training with different backgrounds yet.

How do I use masks during trianing?

ruiqiyan commented 2 weeks ago

Hi, currently the background is set to random by default. Can you try adding the flag--pipeline.model.background-color whitewhen training withns-train dn-splatter [options] --pipeline.model.background-color white --data path_to_data/this will set the background color to white. One other option if this does not work is to use masks while trianing. Let me know if any of these help. You can also post images/screenshots so I can see the results better. I have not tried training with different backgrounds yet.

How do I use masks during trianing?

Do you use the format of the DTU dataset?