rogermm14
commented
1 year ago Hello @wuy202 ! I am not sure I understood your comment. Rays intersecting the same geographic point are casted from different viewing angles according to the RPC models and therefore contribute to different image pixels. There are no repreated rays unless two images share the same exact viewing direction and RPC model. Best, Roger
Given that ray origins and destinations are computed under ECEF, it is likely that rays crossing the same geographic point would have the same coordinates regardless of which RPCs are involved in computing. I assume this will dramatically decrease the number of unique rays available in training and make Sat-Nerf less likely to be able to synthesize novel views.