dreamgaussian
Mesh generation from 3DGS output format
@hodakagoto Hi, yes! You can use the load config to directly load the ply files.
@ashawkey I got an error when I ran this command:
python main.py --config configs/image.yaml load=/path/to/3dgs-output/input.ply
Am I doing something wrong?
I also tried this command but got an error:
python main.py --config configs/image.yaml load=miku.ply gui=True
Number of points at loading : 531242
Traceback (most recent call last):
File "D:\work\dreamgaussian\main.py", line 915, in <module>
gui = GUI(opt)
File "D:\work\dreamgaussian\main.py", line 78, in __init__
self.renderer.initialize(self.opt.load)
File "D:\work\dreamgaussian\gs_renderer.py", line 713, in initialize
self.gaussians.load_ply(input)
File "D:\work\dreamgaussian\gs_renderer.py", line 438, in load_ply
assert len(extra_f_names)==3*(self.max_sh_degree + 1) ** 2 - 3
AssertionError
What should I do?
maybe you should rename the directory of '3dgs-output' to '3dgs_output' and change the code to "python main.py --config configs/image.yaml load=/path/to/3dgs_output/input.ply"
@maybeLx The path is an example and does not actually contain -. I get the same error as @camtrik did.
I also tried this command but got an error:
python main.py --config configs/image.yaml load=miku.ply gui=True Number of points at loading : 531242 Traceback (most recent call last): File "D:\work\dreamgaussian\main.py", line 915, in <module> gui = GUI(opt) File "D:\work\dreamgaussian\main.py", line 78, in __init__ self.renderer.initialize(self.opt.load) File "D:\work\dreamgaussian\gs_renderer.py", line 713, in initialize self.gaussians.load_ply(input) File "D:\work\dreamgaussian\gs_renderer.py", line 438, in load_ply assert len(extra_f_names)==3*(self.max_sh_degree + 1) ** 2 - 3 AssertionErrorWhat should I do?
From the code, the sh_degree should be 3 for personal given .ply, but need to be 0 for DreamGaussian trained name_model.ply. For the former one, the mesh quality is very bad, can not see what it is...
@WWmore The visualize command could work with sh_degree = 0. But I don't know how to generate a mesh. Once I run this command python main.py --config configs/image.yaml load=point_cloud.ply, got an error:
Traceback (most recent call last):
File "/dreamgaussian/main.py", line 882, in <module>
gui.train(opt.iters)
File "/dreamgaussian/main.py", line 858, in train
self.train_step()
File "/dreamgaussian/main.py", line 237, in train_step
loss.backward()
AttributeError: 'int' object has no attribute 'backward'
@hodakagoto I think we can not train it (since the loss==0 here), but directly extract_mesh to get a mesh. I tried one example in the GUI window as seen in my note here https://github.com/WWmore/dreamgaussian. Unfortunately, the mesh quality is very bad. And I am also trying to figure out how to get a good mesh with nice texture from given Gaussian Splatting .ply. Maybe the authors or others can solve such problem easier and sooner. Looking forwards to the post.
/dc2/dreamgaussian$ python main.py --config configs/image.yaml load=/dc2/SuGaR/output/0001_matting_blackbg/point_cloud/iteration_7000/point_cloud.ply
Number of points at loading : 81269
Traceback (most recent call last):
File "/dc2/dreamgaussian/main.py", line 921, in
same issue here!!!
@hodakagoto I think we can not train it (since the loss==0 here), but directly extract_mesh to get a mesh. I tried one example in the GUI window as seen in my note here https://github.com/WWmore/dreamgaussian. Unfortunately, the mesh quality is very bad. And I am also trying to figure out how to get a good mesh with nice texture from given Gaussian Splatting .ply. Maybe the authors or others can solve such problem easier and sooner. Looking forwards to the post.
老哥，你明白model.ply对于导出obj有什么作用了吗，在哪调用model.ply了
AttributeError: 'int' object has no attribute 'backward' How to solve the problem？