Open sx123456123 opened 1 year ago
Thanks for your good question.
Generally speaking, our proposed framework could work under moving cameras since the original NeRF was designed for the multi-view setting. But the performance might compromise when the input camera poses are not accurate enough or the input images contain many artifacts, like motion blur. Also, you may need to modify some training options, e.g., disabling NDC.
You put the camera in a fixed single view point, if my camera moves, does this code still work effectively?