bounding box issue

[WWaynee]

trafficstars

Thanks for your amazing job, I love this project the way I love you.

【I finished installing conda environment and tried training mipnerf360 dataset, then a error occured:】

Initialize 3DGS using /dataset/mipnerf360/bicycle/sparse/0/points3D.ply Traceback (most recent call last): File "/project/DOGS/train.py", line 72, in train(local_config) File "/DOGS/train.py", line 25, in train trainer = create_trainer(config) File "/project/DOGS/utils.py", line 17, in create_trainer trainer = GaussianSplatTrainer(config, prefetch_dataset, trainset, valset, model) File "/project/DOGS/conerf/trainers/gaussian_trainer.py", line 138, in init super().init(config, prefetch_dataset, File "/project/DOGS/conerf/trainers/implicit_recon_trainer.py", line 80, in init self.build_networks() File "/project/DOGS/conerf/trainers/gaussian_trainer.py", line 186, in build_networks self.init_gaussians() File "/project/DOGS/conerf/trainers/gaussian_trainer.py", line 173, in init_gaussians self.bounding_box = torch.tensor(bounding_box, dtype=torch.float32) TypeError: must be real number, not NoneType

What's wrong with this?

【And another question about this line 23 in scripts/env/install.sh:】 conda remove ffmpeg --force

Why should ffmpeg be removed? I tried this and an error occured when importing cv2: ImportError: libavcodec.so.57: cannot open shared object file: No such file or directory As far as I know libavcodec.so is something related to ffmpeg.

[AIBluefisher]

Hi, the issue has been fixed. Please use the latest update of the admm branch for the experiment.

The conda remove ffmpeg --force command is used since the system ffmpeg may conflict with the conda installed ffmpeg. If you do not have this issue, just ignore this command.

[AIBluefisher]

Please refer to https://github.com/AIBluefisher/DOGS/issues/10 for the same issue.

[coder12333444]

@AIBluefisher
I use ADMM branch,but I find

I commented out old_diff_gaussian_rasterization and ran an error. I think it is a problem with this library. How to install this library? error: /conerf/trainers/trainer.py", line 76, in __del__ if self.writer is not None: AttributeError: 'GaussianSplatTrainer' object has no attribute 'writer'

[AIBluefisher]

@AIBluefisher I use ADMM branch,but I find

I commented out old_diff_gaussian_rasterization and ran an error. I think it is a problem with this library. How to install this library? error: `/conerf/trainers/trainer.py", line 76, in __del__ if self.writer is not None: AttributeError: 'GaussianSplatTrainer' object has no attribute 'writer'`

You should install it by pip install submodules/old-diff-gaussian-rasterization.

[coder12333444]

@AIBluefisher
Thank you very much, I have successfully run the ADMM branch code on a single node but where is the final result stored? No final ply file (how can I visualize it)? Similar to the results of the original version of gaussian splatting the results :

[AIBluefisher]

Hi, we do not support a final ply file for native rendering. Instead, we provide a web tool for visualization. A splat file is under the eval/val/splats folder. All you need to do is upload the splat file to the web tool and then you can view the 3D model.