I would like to test the project with my own dataset. How should I configure my own configs file? When I train the model without setting the configs file, the SSIM metric is much worse compared to 3DGS, while other metrics are on par with 3DGS. Can you help me with this?

[az396]

No description provided.

[benhenryL]

Hi, thanks for your interest to our work.
In case of motion blur, hyperparameter num_moments should be bigger to model camera movement if your dataset exhibits severe motion blur.
You can also slightly increase lambda_s or max_clamp to address more harsh defocus blur.

Or maybe you can refer to the FAQ of 3D-GS:
... However, the approach can struggle in multi-scale detail scenes (extreme close-ups, mixed with far-away shots). This is usually the case in, e.g., driving data sets (cars close up, buildings far away). For such scenes, you can lower the --position_lr_init, --position_lr_final and --scaling_lr (x0.3, x0.1, ...). The more extensive the scene, the lower these values should be. ...

I can help you more if I can see the dataset, so would you like to share some train/test images and renderings if you don't mind?
Thanks!

[az396]

Dear author,
Thank you very much for your response and suggestions regarding our work. We greatly appreciate your guidance on the hyperparameters and the insights provided in the FAQ section of 3D-GS.
We highly acknowledge your advice on handling motion blur. We will adjust the "num_moments" hyperparameter to better model camera movement and slightly increase "lambda_s" or "max_clamp" to address more severe defocus blur. Additionally, we will consider your suggestion to lower --position_lr_init, --position_lr_final, and --scaling_lr based on the complexity of the scene.
We would also like to express our gratitude for your willingness to provide further assistance and support.