The nerfstudio team has quietly introduced a new feature that has the potential to enhance the quality of video renders. This feature allows users to automatically align their camera path with the perspective of the closest original training image used in the training process.
In order to utilize this new feature, set--render-nearest-camera.
When you activate the --render-nearest-camera” setting in nerfstudio, the software automatically selects the closest matching training image to your desired camera path. nerfstudio evaluates the camera’s position and rotation, then calculates a weighted sum of these distances to find the nearest training image. This method ensures that the chosen perspective is the closest match to your camera path.
You might be wondering, what if that new camera path pulls it through a section that is occluded or has poor visibility? nerfstudio has you covered, adding a togglable --check-occlusions that will check the line of sight. By default, it is disabled.
As a result of this, we can expect to see higher fidelity video renders coming out from users of the Berkeley based platform. nerfstudio continues to be open source, so if you would like to contribute, check out their Github.
There is still an open Pull Request to formally add Gaussian Splatting to the list of supported methods; though to my understanding the delay is not coming from the nerfstudio side. Hopefully we will see it be merged soon.
