You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I produced very poor results when running Gaussian_stlatting, not only on my own dataset, but also on the MipNeRF360 scenestandt_db provided by the training project.
There is a significant gap between my training results and the results in the pre-trained model provided by the project, I don't know where the problem lies, I haven't changed any hyperparameters at all. The hyperparameter settings are the same as on the website:
I don't know if my hardware didn't meet the requirements, but I feel like the result shouldn't be so bad, my hardware is as follows:
GPU: Nvidia RTX2070super 8GB
RAM: 16 GB
My reconstruction results are compared with the pre trained results as follows:
My Result:
Pre-trained:
The observation results showed that there was a significant difference in the initial state of the splats, and as the splashes increased, *many large white spots appeared in the results.
And I also found that there is a significant difference in memory size between point.cly in my iteration30000 and point.cly in pre trained iteration30000:
Not only is there an issue in this particular scenario, but all demo scenarios and my own datasets (train、trucks、playroom……)currently have the same problem, Can someone tell me where the problem lies?
The text was updated successfully, but these errors were encountered:
@o-o-cloud hello,I was lucky enough to solve this problem after sending it out last night,I ran Gaussian Splatting again using Colab and found that the final result was good. I hope this can help you solve this problem.
The problem is likely due to hardware issues with the computer, but I am still unsure why the hardware is affecting the training results
@yuyunlong2002 Thank you for your enthusiastic help, but I think my hardware is enough.
RTX40 series
I trained under ubuntu and visualized under windows
I produced very poor results when running Gaussian_stlatting, not only on my own dataset, but also on the MipNeRF360 scenes
tandt_db
provided by the training project.There is a significant gap between my training results and the results in the pre-trained model provided by the project, I don't know where the problem lies, I haven't changed any hyperparameters at all. The hyperparameter settings are the same as on the website:
I don't know if my hardware didn't meet the requirements, but I feel like the result shouldn't be so bad, my hardware is as follows:
My reconstruction results are compared with the pre trained results as follows:
My Result:
Pre-trained:
The observation results showed that there was a significant difference in the initial state of the splats, and as the splashes increased, *many large white spots appeared in the results.
And I also found that there is a significant difference in memory size between point.cly in my iteration30000 and point.cly in pre trained iteration30000:
Not only is there an issue in this particular scenario, but all demo scenarios and my own datasets (train、trucks、playroom……)currently have the same problem, Can someone tell me where the problem lies?
The text was updated successfully, but these errors were encountered: