mega-nerf icon indicating copy to clipboard operation
mega-nerf copied to clipboard

blurry results?

Open pengzhao-life opened this issue 2 years ago • 2 comments

Hi, I can train model with my custom data, but the results are kind of blurry, i.e. details in the source images are lost. Is it caused by inaccurate camera pose estimation? Thanks!

I forgot to mention. I followed the steps for camera pose estimation, except for the step of 'colmap model_aligner', where argument --database_path /path/to/databse.db, I provided the database.db file from the previous step, but I got SQLite error when loading the database file. So I skipped this step. Maybe this caused blurry? What confused me is that the results still look reasonable, but just blurry.

pengzhao-life avatar Sep 20 '22 22:09 pengzhao-life

@pengzhao-life Hi , i want to run this script https://gist.github.com/hturki/b3ae2e020bcc2a08dbba402786f1fcd3 on the images which i have. what i have understood is i need to install colmap,pycolmap then install pixsfm and hloc.After that i run this script directly on the env to get the output however i am getting below error image

image

Kindly tell me if i am doing anything wrong in installation or it has to be other way round ?

thanks

sumanttyagi avatar Sep 20 '22 22:09 sumanttyagi

Hi @sumanttyagi, I followed this link for installation https://github.com/cvg/pixel-perfect-sfm Then for my case, I need to select python interpreter /bin/python3, rather than other conda environment, in order to import hloc. I am not sure if you have the same issue.

pengzhao-life avatar Sep 20 '22 23:09 pengzhao-life