chiehwangs / gaussian-head Goto Github PK

python .\submodules\depth-diff-gaussian-rasterization\setup.py install

Traceback (most recent call last):
File ".\submodules\depth-diff-gaussian-rasterization\setup.py", line 21, in
CUDAExtension(
File "C:\Users\muham\anaconda3\envs\ghead\lib\site-packages\torch\utils\cpp_extension.py", line 1048, in CUDAExtension
library_dirs += library_paths(cuda=True)
File "C:\Users\muham\anaconda3\envs\ghead\lib\site-packages\torch\utils\cpp_extension.py", line 1186, in library_paths
paths.append(_join_cuda_home(lib_dir))
File "C:\Users\muham\anaconda3\envs\ghead\lib\site-packages\torch\utils\cpp_extension.py", line 2223, in _join_cuda_home
raise EnvironmentError('CUDA_HOME environment variable is not set. '
OSError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root.

Hello I got this working
I ran
python render.py -m ./output/id1
during inference on my 4gb i saw

Rendering progress: 100%|██████████████████████████████████████████████████████████| 2254/2254 [49:41<00:00,  1.32s/it]
Rendering progress: 100%|██████████████████████████████████████████████████████████████| 50/50 [01:33<00:00,  1.87s/it]

Could you explain what is happening during the above part, due to the duration it took I'm wondering if that's a step that does not need to be repeated?

Also I have the modified dataset downloaded and was wondering if I can do cross identity reenactment with it or if I need to do some training on those persons first?

Hi Jie,

Many thanks for your awesome job!

May I inquire about the training time and the corresponding GPU device you using?

Any plans to add a license to the repo?

when I use render.py,Looking for config file in ./output/id1/cfg_args
Config file found: ./output/id1/cfg_args
Rendering ./output/id1/
Traceback (most recent call last):
File "render.py", line 97, in
render_sets(model.extract(args), args.iteration, pipeline.extract(args), args.warm_up, args.is_debug, args.skip_train, args.skip_test,
File "render.py", line 48, in render_sets
scene = Scene(dataset, gaussians, is_debug=is_debug, novel_view=novel_view, only_head=only_head, load_iteration=iteration, shuffle=False)
File "/data/liukean/projects/gaussian-head/scene/init.py", line 25, in init
self.loaded_iter = searchForMaxIteration(os.path.join(self.model_path, "point_cloud"))
File "/data/liukean/projects/gaussian-head/utils/system_utils.py", line 29, in searchForMaxIteration
saved_iters = [int(fname.split("")[-1]) for fname in os.listdir(folder)]
File "/data/liukean/projects/gaussian-head/utils/system_utils.py", line 29, in
saved_iters = [int(fname.split("")[-1]) for fname in os.listdir(folder)]
ValueError: invalid literal for int() with base 10: 'Store'

I didn't see a direct interface for reenactment, but I did see it in the project demonstration. Where is this feature located in the code?

Hi, thanks for open-sourcing this good work. How to apply this method to a custom dataset? Could you provide the data processing code?

as title, the train.py file is not appearing in this git

hey thank you for sharing this wonderful work. i got an error when running python render.py -m ./output/id1

and the \output\id1\point_cloud\iteration_80000 point_cloud is a bee, is that the right pretrained model?

Hi, thank you for your nice work.
When will you release the training code?

Congrats on being the first person to upload code related to Gaussian head avatars, you even beat your timeline for release so that was awesome!

I have a couple of points,
the first being the link to the project page on the repo links back to the repo I would be good to see a video example if a project page does exist. Edit: I just noticed that's on your to-do list so you can ignore this part.

The second point I would like to bring up is
I have installed everything correctly and when I run in Windows annaconda
I get the following error message:

(gaussian-head) C:\Users\Jack\gaussian-head>python render.py -m ./output/id1
Looking for config file in ./output/id1\cfg_args
Config file found: ./output/id1\cfg_args
Rendering ./output/id1
Loading trained model at iteration 80000 [03/01 16:10:37]
C:\Users\Jack\gaussian-head\data\id1 [03/01 16:10:37]
Traceback (most recent call last):
  File "render.py", line 97, in <module>
    render_sets(model.extract(args), args.iteration, pipeline.extract(args), args.warm_up, args.is_debug, args.skip_train, args.skip_test,
  File "render.py", line 48, in render_sets
    scene = Scene(dataset, gaussians, is_debug=is_debug, novel_view=novel_view, only_head=only_head, load_iteration=iteration, shuffle=False)
  File "C:\Users\Jack\gaussian-head\scene\__init__.py", line 36, in __init__
    assert False, "Could not recognize scene type!"
AssertionError: Could not recognize scene type!

Hi, when I use your model for training, there report an error:
The size of tensor a (9) must match the size of tensor b (3) at non-singleton dimension 1

I wonder if this line of code may be changed from if input.shape[1] != 3: to if input.shape[0] != 3:?

I really hope you can give me an answer, thank you!

First, thanks for your great work, and I'd like to know the the specific setting for the trainset and the testset of each id?

As the paper mentioned:

The training data we use for each subject consists of approximately 2000 frames, and evaluate the results using the last 200 frames.

While the number of every id is range of 3000~4200 frames. is the trainset the first 2000 frames and the last 200 frames of the datas as the testset?

Hello@chiehwangs, thank you for your contribution to the community. Can you provide other pre-trained models for face ID? This will be of great help to our current work.

How long the infering time is？

Excuse me. I wonder how to install submodules\simple-knn and depth-diff-gaussian-rasterization in requirments.txt.
Expecting your response.

How to create a dataset of my head for training？

Hello, I want to try to render different head pose orientations, but when I try to render the pre-trained model you provided using each of the following three commands, I get the same result head orientation, can you give me some suggestions?
python render.py --novel_view 5 -m ./output/id1
python render.py --novel_view 30 -m ./output/id1
python render.py --novel_view 50 -m ./output/id1

Thanks for sharing this great work. Hey, just wondering under what conditions you might consider open sourcing the train.py file? I'm really looking forward to being able to train the model on my own dataset.