Closed chenkang455 closed 2 months ago
Hello, thanks for your interest in our work.
BAD-Gaussians is the first open-source project that provides motion-deblurring for 3DGS. It is aimed to become the successor of BAD-NeRF, for its much faster training speed, real-time rendering capability and superior performance.
This code base was built on top of our previous open-source project that re-implements BAD-NeRF in the
nerfstudio
framework. The core code has been completely refactored compared to BAD-NeRF, and we tried our best to keep the code base simple and easy to follow, following the the style guides of the community.
Thanks a lot for your detailed response!
Hi, @LingzheZhao sry to bother you again. I am curious about the performance of BAD-Gaussian on the 360 scene like lego. Can the BAD-Gaussian converge nicely on the 360 scene?
Thx a lot!
Hi, @chenkang455, yes it can, because 3DGS itself works nicely on 360 (object-centric) scenes, unlike original BAD-NeRF was designed for LLFF datasets with NDC scene contraction.
In fact, lots of NeRF-based implicit methods, need special modeling of scene contraction, while 3DGS-based methods usually do not.
Hi, @chenkang455, yes it can, because 3DGS itself works nicely on 360 (object-centric) scenes, unlike original BAD-NeRF was designed for LLFF datasets with NDC scene contraction.
In fact, lots of NeRF-based implicit methods, need special modeling of scene contraction, while 3DGS-based methods usually do not.
I get it, thanks for your detailed response again. 😊
Hello, thanks for your interest in our work.
BAD-Gaussians is the first open-source project that provides motion-deblurring for 3DGS. It is aimed to become the successor of BAD-NeRF, for its much faster training speed, real-time rendering capability and superior performance.
This code base was built on top of our previous open-source project that re-implements BAD-NeRF in the
nerfstudio
framework. The core code has been completely refactored compared to BAD-NeRF, and we tried our best to keep the code base simple and easy to follow, following the the style guides of the community.