extract_mesh_tsdf.py error

[zhouilu]

command is OMP_NUM_THREADS=4 CUDA_VISIBLE_DEVICES=0 python extract_mesh_tsdf.py -s data -m output/8a06585b-5/ --eval -i images --iteration 7000 --voxel_size 0.004 --sdf_trunc 0.04 -r 4 --data_device=cpu --sh_degree 0 get error, here is log

Reading camera 463/463
Loading Training Cameras
Loading Test Cameras
Loading Ply from output/8a06585b-5/point_cloud/iteration_7000/point_cloud.ply
Number of points at initialisation :  1016171
Rendering progress: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 405/405 [00:05<00:00, 73.97it/s]
[Open3D WARNING] Write PLY failed: mesh has 0 vertices.
mesh saved at output/8a06585b-5/tsdf/ours_7000/mesh.ply
post processing the mesh to have 1 clusterscluster_to_kep
[Open3D DEBUG] [ClusterConnectedTriangles] Compute triangle adjacency
[Open3D DEBUG] [ClusterConnectedTriangles] Done computing triangle adjacency
[Open3D DEBUG] [ClusterConnectedTriangles] Done clustering, #clusters=0

i use 0 sh_degree, and tune.py render.py metrics.py are running ok. could you give some advice? thanks~

[YuxueYang1204]

Hi @zhouilu ! Have you ever tried visualizing the Gaussian model with sh_degree=0? The mesh extraction is based on the rendered depth, so it won't fail as long as the depth map can be rendered. You can check it with python render.py -m {$output_path} --iteration 7000 --render_other (Trim3DGS) or python render.py -m {$output_path} --iteration 7000 --skip_mesh (Trim2DGS). The rendered maps are stored in $output_path/train/ours_7000. Hope this helps!

[zhouilu]

sh_degree=0 render is ok in https://playcanvas.com/supersplat/editor, and got good psnr..

[YuxueYang1204]

Are the depth maps rendered successfully?

sh_degree=0 render is ok in https://playcanvas.com/supersplat/editor, and got good psnr..

[zhouilu]

Are the depth maps rendered successfully?

sh_degree=0 render is ok in https://playcanvas.com/supersplat/editor, and got good psnr..

yes. like

[YuxueYang1204]

If possible, could you upload an output sample that includes point_cloud/iteration_xxx/point_cloud.ply and cfg_args? I'm willing to analyze the specific issue based on it.