“out of memory” when running extract_mesh.py

[Nabla-GIT]

hello，
I'm running extract_mesh.py to extract the mesh from 15000.pt:

python extract_mesh.py -s gaussian_splatting\datas\inputs\data1\ -c gaussian_splatting\datas\outputs\out1\ -m output\coarse\data1\sugarcoarse_3Dgs7000_densityestim02_sdfnorm02\15000.pt

but it turns out: out of memory!!!(see details below)

VRAM of my GPU(rtx 4050 laptop) is 8GB, I know it is insufficient

what should I do?

Thanks a lot !

details:

-----Parameters-----
Source path: gaussian_splatting\datas\inputs\data1\
Gaussian Splatting Checkpoint path: gaussian_splatting\datas\outputs\out1\
Coarse model Checkpoint path: output\coarse\data1\sugarcoarse_3Dgs7000_densityestim02_sdfnorm02\15000.pt
Mesh output path: output\coarse_mesh\data1
Surface levels:
[0.1, 0.3, 0.5]
Decimation targets:
[200000, 1000000]
Use custom bbox: False
Use eval split: True
GPU: 0
Use centers to extract mesh: False
Use marching cubes: False
Use vanilla 3DGS: False
--------------------
Loading the initial 3DGS model from path gaussian_splatting\datas\outputs\out1\...
Found image extension .JPG
9 training images detected.
The model has been trained for 7000 steps.

Loading the coarse SuGaR model from path output\coarse\data1\sugarcoarse_3Dgs7000_densityestim02_sdfnorm02\15000.pt...
Use min to initialize scales.
Initialized radiuses for 3D Gauss Rasterizer
Coarse model loaded.
Coarse model parameters:
_points
torch.Size([213781, 3])
True
all_densities
torch.Size([213781, 1])
True
_scales
torch.Size([213781, 3])
True
_quaternions
torch.Size([213781, 4])
True
_sh_coordinates_dc
torch.Size([213781, 1, 3])
True
_sh_coordinates_rest
torch.Size([213781, 15, 3])
True
Number of gaussians: 213781
Opacities min/max/mean: tensor(9.6427e-05, device='cuda:0') tensor(1., device='cuda:0') tensor(0.6500, device='cuda:0')
Quantile 0.0: 9.642718214308843e-05
Quantile 0.1: 0.0025188217405229807
Quantile 0.2: 0.01307610608637333
Quantile 0.3: 0.31468358635902405
Quantile 0.4: 0.727897584438324
Quantile 0.5: 0.9511145949363708
Quantile 0.6: 0.9998317956924438
Quantile 0.7: 0.9999556541442871
Quantile 0.8: 0.9999804496765137
Quantile 0.9: 0.9999923706054688

Starting pruning low opacity gaussians...
WARNING! During optimization, you should use a densifier to prune low opacity points.
This function does not preserve the state of an optimizer, and sets requires_grad=False to all parameters.
Number of gaussians left: 143904
Opacities min/max/mean: tensor(0.5000, device='cuda:0') tensor(1., device='cuda:0') tensor(0.9345, device='cuda:0')
Quantile 0.0: 0.5000038146972656
Quantile 0.1: 0.7125014066696167
Quantile 0.2: 0.8746340274810791
Quantile 0.3: 0.9927117824554443
Quantile 0.4: 0.9998162388801575
Quantile 0.5: 0.9999368190765381
Quantile 0.6: 0.9999655485153198
Quantile 0.7: 0.999980092048645
Quantile 0.8: 0.9999889135360718
Quantile 0.9: 0.9999951124191284
Processing frame 0/9...
Current point cloud for level 0.1 has 0 points.
Current point cloud for level 0.3 has 0 points.
Current point cloud for level 0.5 has 0 points.
Traceback (most recent call last):
  File "C:\Users\15088\Desktop\Short PHD 3\3D Gaussian\sugar1\SuGaR-main\extract_mesh.py", line 46, in <module>
    extract_mesh_from_coarse_sugar(args)
  File "C:\Users\15088\Desktop\Short PHD 3\3D Gaussian\sugar1\SuGaR-main\sugar_extractors\coarse_mesh.py", line 277, in extract_mesh_from_coarse_sugar
    frame_surface_level_outputs = sugar.compute_level_surface_points_from_camera_fast(
  File "C:\Users\15088\Desktop\Short PHD 3\3D Gaussian\sugar1\SuGaR-main\sugar_scene\sugar_model.py", line 1995, in compute_level_surface_points_from_camera_fast
    closest_gaussian_inv_scaled_rotation = gaussian_inv_scaled_rotation[pass_closest_gaussians_idx]
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1.07 GiB (GPU 0; 6.00 GiB total capacity; 10.50 GiB already allocated; 0 bytes free; 11.29 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF