Whether dynibar has the ability to render larger time spans in fixed viewing angles？

[sjdfkjawe]

I noticed that in the girl in green clothes dataset, you render 74 moments from a fixed perspective. I reproduced the experiment and found that the time span in a fixed view was around 70 frames.
I see the quality of the first few frames and the last few frames of the render results is not very good, of course, the further away from the training views, the worse the quality.

I would like to ask you if dynibar has any hyperparameters that can be adjusted to render more moments from fixed vies, how to get more than 74 moments when render in a fixed view of the girl in green dress dataset.
Is dynibar unable to render plenty moments images under fixed viewing angles.

[sjdfkjawe]

your render results from dynibar project page:
time 0:

time 45:

time 73:

my reproduced results, I choose the the time 51 background as the fixed render view:
time 0:

time 45:

time 74

[zhengqili]

Hi I think due to limited training supervision signal, current IBRNet can only rendered limited viewpoints and time ranges (meaning in this case, if girl is very further away from original viewpoint or time, the quality would degrade very quickly)

[CTouch]

I only found the datasets from the NSFF, and there is no girls_in_green_clothe. Could you please tell me where to find the data?
Thank you !

[sjdfkjawe]

I only found the datasets from the NSFF, and there is no girls_in_green_clothe. Could you please tell me where to find the data? Thank you !

Sorry for waiting a long time, you can get this dataset from https://zhmeishi.github.io/dvs/