Slow open3d mesh extraction

[NIRVANALAN]

Hi! Thanks for the wonderful work. Regarding mesh extraction, after training your demo scene (m360 garden), I run the demo code python render.py -s ../3dgs/dataset/garden -m output/m360/garden/ --skip_test --skip_train --mesh_res 1024 for mesh extraction. However, it seems really slow (with high CPU usage). while after 30 minutes, the TSDF integration goes only 19 iterations. I wonder whether this is reasonable since m360 scene is extremely large, or I didn't get open3d installed right (looks like the TSDF integration is working on the CPU). Any suggestions?

Rendering output/m360/garden/                                                                                                                                                                 [0/2613]
Loading trained model at iteration 30000                                                                                                                                                              
Reading camera 185/185                                                                                                                                                                                
Loading Training Cameras                                                                                                                                                                              
[ INFO ] Encountered quite large input images (>1.6K pixels width), rescaling to 1.6K.                                                                                                                
 If this is not desired, please explicitly specify '--resolution/-r' as 1                                                                                                                             
Loading Test Cameras                                                                                                                                                                                  
export mesh ...                                                                                                                                                                                       
reconstruct radiance fields: 185it [00:03, 47.08it/s]                                                                                                                                                 
The estimated bounding radius is 3.28                                                                                                                                                                 
Use at least 6.57 for depth_trunc                                                                                                                                                                     
Running tsdf volume integration ...                                                                                                                                                                   
voxel_size: 0.006414120935716085                                                                                                                                                                      
sdf_trunc: 0.032070604678580424                                                                                                                                                                       
depth_truc: 6.568059838173271                                                                                                                                                                         
TSDF integration progress: 19it [34:27, 110.53s/it]

[hbb1]

Hi, algorithm-wise, TSDF is very computationally efficient. By default we use a CPU version of TSDF but it should typically take less than 2 minutes for m360. You should make sure correct installation or you have sufficient CPU cores. Another future improvement is to use a GPU version of TSDF.

[hirotong]

Hi, I also met the problem, try to limit the number of CPU cores used by OMP_NUM_THREADS=4.

[NIRVANALAN]

Cool, let me give it a try on my side.

[hjh530]

Did it work？

Cool, let me give it a try on my side.