Question about rendering results of trained model.

[Kim-GeonU]

Thank you for your great work and open-sourcing your code for the community.

I tried training the sequence 114 in the Waymo open dataset by following your instructions, but the trained results seem to have problems.

Above results are saved using following code, which just saves images from render_results in do_evaluation() function. I used 30,000 iterations model.

And I didn't change any other code except above visualization code.

The above rendering results seem to fail to render dynamic objects and also fail to distinguish between static and dynamic objects. Do you have any idea what causes this problem?

These are logs during training.

Thank you!

[XuHu0529]

I have the same results. It seems to fail to reconstruct the dynamic objects.

[tapowanliwuyun]

Thank you for your great work and open-sourcing your code for the community. I have the same results.
Does this mean that the configuration parameters downloaded need to be modified in order to achieve the desired results?

[nnanhuang]

Thank you for your great work and open-sourcing your code for the community.

I tried training the sequence 114 in the Waymo open dataset by following your instructions, but the trained results seem to have problems.

Above results are saved using following code, which just saves images from render_results in do_evaluation() function. I used 30,000 iterations model. And I didn't change any other code except above visualization code.

The above rendering results seem to fail to render dynamic objects and also fail to distinguish between static and dynamic objects. Do you have any idea what causes this problem?

These are logs during training.

Thank you!

Hello! This sequence does not appear in the two main subsets we use, static32 and dynamic32, and our experimental data in the paper align with both emernerf and street gaussian standards. As such, we have not previously encountered this issue. You might consider experimenting with different parameters, for example, reducing the number of training frames. We will investigate this further and study this particular scenario more closely. Thank you for your interest!

[nevergone123]

I got the similar bad results using scene 754, which is included in the dynamic 32 splits.
Below is the dynamic_rgb:

Do you have any idea to fix it? Many thanks！ @nnanhuang

[Kim-GeonU]

I also observed some artifacts in a rendered image from the scene 003 included in the static 32 splits.