Closed Caquita-beep closed 5 months ago
Hi, training, testing and preprocessing are tested on Ubuntu 20.04. More efforts are needed for windows. (like slash in windows..). I recommend that you can use Ubuntu for training. (scripts use Ubuntu file path convention )
Hello. Yeah I noticed. Had errors with pre_n3d.py becuase of the "slash" and as soon as I changed to / it worked hehe but I guess it didn't as it's giving the wrong output...
Hello. Yeah I noticed. Had errors with pre_n3d.py becuase of the "slash" and as soon as I changed to / it worked hehe but I guess it didn't as it's giving the wrong output...
you may need be carefull with other parts like...
and somecodes in the https://github.com/oppo-us-research/SpacetimeGaussians/blob/2f56210489ca115dc006c3400cc54750b2d38622/thirdparty/gaussian_splatting/helper3dg.py#L181
Hello. Turns out for some reason Colmap installation with conda is having issues and so the "getcolmapsinglen3d" function in helper3dg.py was exiting prematurelly. So I run colmap from another env (which is not installed in the env lol but on the machine) and colmap is wokring this time. However, I'm getting error on triangulation :
F20240104 14:56:36.840480 42368 sfm.cc:473] Check failed: bundle_adjuster.Solve(&reconstruction) Check failure stack trace: @ 00007FFAB66F6E5B public: void cdecl google::LogMessage::Flush(void) ptr64 @ 00007FFAB66F5739 public: cdecl google::LogMessageFatal::~LogMessageFatal(void) ptr64 @ 00007FF64F621E37 (unknown) @ 00007FF64F621360 (unknown) @ 00007FF64F5D65E6 (unknown) @ 00007FF64F5D7DEB (unknown) @ 00007FF64FBC43B4 (unknown) @ 00007FFADA4A257D BaseThreadInitThunk @ 00007FFADABEAA58 RtlUserThreadStart
Any clues?
Hello,
Training, Testing and preprocessing have been tested on Ubuntu 20.04. While our efforts have been focused on this system, much more efforts are need for other system.
We recommend that you use colmap with conda-forge.
For compatibility with Windows, we are busy dealing with other issues, and will try to adapt to Windows later. Please stay tuned for that.
Hello. Yeah I noticed. Had errors with pre_n3d.py becuase of the "slash" and as soon as I changed to / it worked hehe but I guess it didn't as it's giving the wrong output...
you may need be carefull with other parts like...
and somecodes in the
def convertdynerftocolmapdb(path, offset=0):
originnumpy = os.path.join(path, "poses_bounds.npy")
video_paths = sorted(glob.glob(os.path.join(path, 'cam*.mp4')))
projectfolder = os.path.join(path, "colmap_" + str(offset))
#sparsefolder = os.path.join(projectfolder, "sparse/0")
manualfolder = os.path.join(projectfolder, "manual")
# if not os.path.exists(sparsefolder):
# os.makedirs(sparsefolder)
In this code block, is the commented-out part the reason why the sparse is not found?
No, the sparse folder will be created by colmap
If colmap
cannot produce the sparse model, the sparse folder will not be found.
what is your python version for colmap ?
No, the sparse folder will be created by
colmap
If
colmap
cannot produce the sparse model, the sparse folder will not be found.what is your python version for colmap
It's the same as yours. python 3.8.18, pytorch 1.12.1, cudatoolkit 11.6.0.
First you can confirm that colmap is installed with
colmap -h
if there is issue with mpir
# for windows run colmap
conda install conda-forge::mpir
after that, you can change the
code https://github.com/oppo-us-research/SpacetimeGaussians/blob/a977e11cc52a56ac670046049c9d76c151000984/script/pre_n3d.py#L79
to
imagesavepath = os.path.join(savedir, os.path.basename(os.path.dirname(folder)) + ".png")
the command in the command prompt should be like
python script\pre_n3d.py --videopath C:\Users\xxxx\coffee_martini
python should handle the cross platform.
Thank you for your reply!
But this problem still exists. It did not generate the sparse folder.
Otherwise, python can use
python script/pre_n3d.py --videopath ../xxxx/cook_spinach
for windows.
python script\pre_n3d.py --videopath C:\Users\xxxx\coffee_martini
can you try this one
python script\pre_n3d.py --videopath C:\Users\xxxx\coffee_martini
as python will handle the slash
But it still hasn't generated the sparse folder.
But it still hasn't generated the sparse folder.
1) can you remove all the colmap_x folders and rerun the code ? (also I recommend recloen the repo and only change one line that i posted )
2) can you help me verify that each input
folder under the colmap_x
folder contains the same number of frames as the number of cameras ?
We have an update for WSL (Windows Subsystem for Linux) available for Windows users. If you encounter any issues with WSL, please feel free to reopen or open an issue.
Hello
When running train.py, I'm getting:
Traceback (most recent call last): File "./thirdparty/gaussian_splatting\scene\dataset_readers.py", line 768, in readColmapSceneInfo cam_extrinsics = read_extrinsics_binary(cameras_extrinsic_file) File "./thirdparty/gaussian_splatting\scene\colmap_loader.py", line 175, in read_extrinsics_binary with open(path_to_model_file, "rb") as fid: FileNotFoundError: [Errno 2] No such file or directory: 'C:\SpacetimeGaussians\data\cut_roasted_beef\colmap_0\sparse/0\images.bin'
Same with "sparse/0/images.txt"
The colmap_0 (and the rest of) folders r generated by pre_n3d.py. There is no "0" folder within colmap_0 and "sparse" folders is generated within distorted and it's empty.
Not sure what is going on. Any ideas?