search

autonomousvision / sdfstudio

A Unified Framework for Surface Reconstruction
Apache License 2.0
1.94k stars 182 forks source link

can it support RGB images? #43

Open YLongJin opened 1 year ago

YLongJin commented 1 year ago

I use process_nerfstudio_to_sdfstudio.py to process nerfstudio data, nert use extract_monocular_cues.py but the images size is fix,can it self-adaption my data size? and then I skip the step, run train, got bad result.can it support RGB images?

niujinshuchong commented 1 year ago

Hi, you can train the model with only RGB images but the monocular cues helps in indoor scenes in general. Which dataset did you try?

YLongJin commented 1 year ago

I try the data which by iphone in indoor scenes, and then use nerfstudio to convert the format and use process_nerfstudio_to_sdfstudio.py.

krahets commented 1 year ago

Hi @YLongJin , if you want to keep the original image size, please DO NOT turn on the --mono-prior.

When enabling --mono-prior, the images will be resized to 384*384. Also, you can adjust --crop-mult to scale the images, but the 1:1 ratio is kept.

YLongJin commented 1 year ago

Hi@krahets，I can not get good mesh from iphone images. I think 348*348 is not crux of the problem](javascript:;)