colmap.sh for hypernerf dataset about 4dgaussians HOT 9 CLOSED

I've missed a single line in colmap.sh
I've solved it by changing export CUDA_VISIBLE_DEVICES=1 to export CUDA_VISIBLE_DEVICES=0 as I was using a single GPU.
Thanks!

from 4dgaussians.

Hi,
The code is based assumption that the poses are known, so it's not necessary to mask the dynamic part. Meanwhile, I think leaving the dynamic part in the scene can also let the scripts generate some point clouds to inspire 4DGS's training.
And is that the data file is right? My data download script is
wget https://github.com/google/hypernerf/releases/download/v0.1/vrig_chicken.zip

from 4dgaussians.

Thank you for your comments, I was using the same data as you provided. Still the point_triangulator fails.. What should I check? Colmap files (cameras.txt and images.txt) are built properly.

Besides, in database.py,
assert camera_id == idList[i]
gives an assertion error.

from 4dgaussians.

Hi, can you provide the full logs when executing my scripts?
I reproduce the full process from another computer but it indeed can work orz
btw, can you generate the dense sfm point from anyother colmap format dataset? I.e:

workdir=“data/nsff/Balloon1” # or any other colmap format dataset
echo $workdir
mkdir -p $workdir/dense/workspace
colmap image_undistorter --image_path $workdir/images_colmap --input_path $workdir/sparse/0 --output_path $workdir/dense/workspace
colmap patch_match_stereo --workspace_path $workdir/dense/workspace --workspace_format COLMAP --PatchMatchStereo.geom_consistency true
# colmap stereo_fusion --workspace_path $workdir/dense/workspace --output_path $workdir/dense/workspace/fused.ply

If it indeed doesn’t work, I’ll upload my generated point clouds

from 4dgaussians.

Hi, the test colmap using nsff dataset works. I've downloaded the hypernerf vrig-chicken dataset to ../data/hypernerf. So I use the following script

cd 4DGaussians
cd ../data/hypernerf
wget https://github.com/google/hypernerf/releases/download/v0.1/vrig_chicken.zip
unzip vrig_chicken
cd ../../4DGaussians
bash colmap.sh ../data/hypernerf/vrig-chicken hypernerf

This script generates empty points3D.txt file in sparse_custom folder.

• While running exhaustive_matcher, CUDA error at ../src/colmap/util/cuda.cc:59 - no CUDA-capable device is detected occurs, I think that is the reason for the error in colmap point_triangulator.

from 4dgaussians.

OMG thank you!
I've fixed it in the latest version.

from 4dgaussians.

Thank you for your comments, I was using the same data as you provided. Still the point_triangulator fails.. What should I check? Colmap files (cameras.txt and images.txt) are built properly.

Besides, in database.py, assert camera_id == idList[i] gives an assertion error.

Hi everyone, I also encountered this assertion error when run colmap.sh. Any idea to solve it?

from 4dgaussians.

I am using the latest code. My command is bash colmap.sh data/hypernerf/3dprinter hypernerf. I had 138 images to perform feature_extractor and the results seemed to be normal. This error happened when run python database.py --database_path $workdir/colmap/database.db --txt_path $workdir/colmap/sparse_custom/cameras.txt. Could you please give me some suggestions? Thanks! @mkjjang3598 @guanjunwu

from 4dgaussians.

Hi @MasterBin-IIAU!
To avoid assertion error, I have changed hypernerf2colmap.py file following the real colmap style.

I changed this print(idx,"SIMPLE_PINHOLE",image_size[0]/2,image_size[1]/2,cam['focal_length']/2,cam['principal_point'][0]/2,cam['principal_point'][1]/2,file=object_cameras_file) to print(idx+1,"SIMPLE_PINHOLE",image_size[0]/2,image_size[1]/2,cam['focal_length']/2,cam['principal_point'][0]/2,cam['principal_point'][1]/2,file=object_cameras_file)

You might verify it by running the colmap (in gui using the same data) and checking cameras.bin and images.bin files.

from 4dgaussians.