Closed zjcs closed 1 month ago
Thanks for your issue. In our experiments, the nvs results of rotating at a certain angle was acceptable, but the lane lines would be terrible after translation. Excitingly this has been resolved in our follow-up work! best
I am interested to how you resolve this issuse. In my view, it seems to be a geometry issuse. If possible, can you share some details about that. Otherwise, do you have any plan to write papers or open the code?
We're very sorry, but this method is still an ongoing project, and more information will follow, so feel free to follow our progress! best
As mention in previouse works, 3DGS tend to overfit the training view while fail the new view synthesis. How about street-gaussians in the NVS, such as 2m translation or 45 rotation? Is there many holes and floaters?