Global Model Visualization

[nchuwssf]

Hi, I trained the global model, how should I visualize it

[perrying]

You can use any visualizer for 3D Gaussian Splatting (e.g., SIBR).

[wuglin]

You can use any visualizer for 3D Gaussian Splatting (e.g., SIBR).

Hi, it's a fascinating work! Same problem I met...
After running build_global_model.py I got a .pth file global_model， what is the next step to visualize it with SIBR-viewer?

[perrying]

You need to convert a .pth file to a .ply file by loading a .pth file such as:

Fed3DGS/gaussian-splatting/eval.py

Line 73 in 9d9ca8d

global_model.set_params(global_params)

and then save it by the save_ply function:

Fed3DGS/gaussian-splatting/scene/gaussian_model.py

Line 325 in 9d9ca8d

def save_ply(self, path):

After that, visualize the saved ply file following official instruction.

[CPuwell]

You need to convert a .pth file to a .ply file by loading a .pth file such as:

Fed3DGS/gaussian-splatting/eval.py

Line 73 in 9d9ca8d

global_model.set_params(global_params)

and then save it by the save_ply function:

Fed3DGS/gaussian-splatting/scene/gaussian_model.py

Line 325 in 9d9ca8d

def save_ply(self, path):

After that, visualize the saved ply file following official instruction.

I want to write a script to convert the file into .ply. But I face some problems about 'import'. It can not find the module I want to import:
File "save_model_as_ply.py", line 3, in
from scene.gaussian_model import GaussianModel
ModuleNotFoundError: No module named 'scene'
And I found there are same problem with all the modules that need to import other module in Fed3DGS. But using the command line I can successfully excute the scripts you write. So, I want to know how to solve this problem?
My script:

import argparse
import torch
from scene.gaussian_model import GaussianModel

def main():
    parser = argparse.ArgumentParser()
    parser.add_argument('--model-path', '-m', required=True, help='Path to the pretrained model')
    parser.add_argument('--ply-path', '-p', required=True, help='Path to save the PLY file')
    args = parser.parse_args()

    model = GaussianModel(sh_degree=2, use_img_feats=True)

    pretrained_params = torch.load(args.model_path)
    model.set_params(pretrained_params)

    model.save_ply(args.ply_path)

if __name__ == '__main__':
    main()

[perrying]

It may be solved by running in the gaussian-splatting directory.

[CPuwell]

Much appreciate! I fixed that problem!
But, when I excute the script it got another error:
RuntimeError: Error(s) in loading state_dict for Sequential:
size mismatch for 4.weight: copying a param with shape torch.Size([48, 64]) from checkpoint, the shape in current model is torch.Size([27, 64]).
May I ask how to fix this problem. And which param caused this problem. I changed the number of client to 2 to shorten the time of trainning since I just want to test.

[perrying]

sh_degree should be 3.

[CPuwell]

sh_degree should be 3.

Thank you a lot!!! I got it.
After that I tried to save .ply file and finally got these

And when I use SIBR to visualize it, it tells me I do not have a config file called "cfg_args":
[SIBR] -- INFOS --: Initialization of GLFW
[SIBR] -- INFOS --: OpenGL Version: 4.6.0 NVIDIA 516.40[major: 4, minor: 6]
[SIBR] ## ERROR ##: FILE C:\projects\gauss2\SIBR_viewers\src\projects\gaussianviewer\apps\gaussianViewer\main.cpp
LINE 140, FUNC main
Could not find config file 'cfg_args' at D:\CPu\Fed3DGS\Fed3DGS\data\Final
Should I use some other module or function to save this file?

[wuglin]

You need to convert a .pth file to a .ply file by loading a .pth file such as:

Fed3DGS/gaussian-splatting/eval.py

Line 73 in 9d9ca8d

global_model.set_params(global_params)

and then save it by the save_ply function:

Fed3DGS/gaussian-splatting/scene/gaussian_model.py

Line 325 in 9d9ca8d

def save_ply(self, path):

After that, visualize the saved ply file following official instruction.

Great! I fixed it. Thanks a lot!