search

mbortolon97 / 6dgs

Code of the paper: 6DGS: 6D Pose Estimation from a Single Image and a 3D Gaussian Splatting Model
Other
131 stars 11 forks source link

The code seems to be incomplete #1

Closed sunhan1997 closed 3 months ago

sunhan1997 commented 3 months ago

Thank you so much for open-source this interesting work. I want to run this code. However, the code seems to be incomplete. Can you give the detailed steps? In addition, the python version also seems to have an impact on the code. I am looking forward to hearing from you. Best wishes.

mbortolon97 commented 3 months ago

Thanks for the interest in our work. "However, the code seems incomplete." => I have just tried to run the code directly from GitHub. It runs without issue on my workstation, meaning all the components are in the right place, and the code is complete. There may be some steps that I forgot to mention in my reproduction instructions. Can you give me further details regarding the error so I can better understand the issue? "The Python version also seems to impact the code." => I am currently using Python 3.10.1 (Compilation of Dec 16, 2021, 17:07:58), but I will be surprised that it does not work on other versions of Python. Hover: I am using typing hints on some parts of the code, and there are a few known issues in the Python version before 3.9 (FYI: I am using Ubuntu 22.04, and Python 3.10.1 should be the standard version provided in the apt package manager). If you can provide a bit more details I will be more than happy to help you further.

sunhan1997 commented 3 months ago

Thank you very much for your quick reply. I'm using python 3.7, which is causing a lot of problems. I will change it to 3.9. In addition, I try to train with 'sh tools/launch_all_tanks_and_temple_training.sh', but it shows: image train.py doesn't seem to contain "--config" and "--datadir"

sunhan1997 commented 3 months ago

This is the data: image and the 'inerf_TensoRF_sampling ' is empty. I am looking forward to hearing from you

mbortolon97 commented 3 months ago

Sorry for the mistake. tools/launch_all_tanks_and_temple_training.sh contain code from IFFNeRF that requires a configuration file. Now, the configuration file is not necessary anymore. I updated the script and the readme with the datasets link that I forgot to insert. Can you try and let me know if it worked?

sunhan1997 commented 3 months ago

Thank you for your reply. I changed python to version 3.10.1, which caused simple-knn and diff-gaussian-rasterization to fail to install. I may need a version of setuptools: image

sunhan1997 commented 3 months ago

I see you've updated a part of the code, but the environment.yaml hasn't, it uses python 3.7. I can implement 3DGS training using python 3.7, but there is a small problem: image I change this path. But with python 3.7, the pretain_eval_attention may not work.

sunhan1997 commented 3 months ago

I have trained it, but the result is not so good: image Maybe there's something wrong during the training ?

sunhan1997 commented 3 months ago

If i use the python 3.10.1, there are more problems: image

mbortolon97 commented 3 months ago

Ah, okay. You were using Conda. I usually stick with virtualenv, as it is generally better. environment.yml was the original one from 3D Gaussian Splatting. I have now updated it, and from my test, should work now.

sunhan1997 commented 3 months ago

Thank you very much for your patient answer, now the code can run normally. If possible, could you provide the "visual.py" to display the prediction results? Thank you again for open-sourcing this excellent work.

mbortolon97 commented 3 months ago

Glad to hear that the code ran. Unfortunately, I cannot release the visual library due to copyright constraints as it was built by various people over the course of multiple years and we were unable to contact some of them. I suggest you to play with a LLM asking it to build this visualisations using Open3D and you should able to replace visual.

On Mon, 29 Jul 2024, 06:51 sunhan, @.***> wrote:

Thank you very much for your patient answer, now the code can run normally. If possible, could you provide the "visual.py" to display the prediction results? Thank you again for open-sourcing this excellent work.

— Reply to this email directly, view it on GitHub https://github.com/mbortolon97/6dgs/issues/1#issuecomment-2254997933, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGCAED6BFHCGHFUO2QTQTLTZOXJ4TAVCNFSM6AAAAABLOKWMUWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENJUHE4TOOJTGM . You are receiving this because you were assigned.Message ID: @.***>

sunhan1997 commented 3 months ago

Ok, thanks for the excellent code anyway😄