Xf = shift_coords extra_xf xf_pos assert np. # NerF will cycle forward, so lets cycle backward. Xf_pos = pos - average_position # barbershop_mirros_hd_dense: # - camera plane is y+z plane, meaning: constant x-values # - cameras look to +x # Don't ask me. ID: generate_transform_matrix( pos, rot): You are receiving this because you are subscribed to this thread.Message Reply to this email directly, view it on GitHub System, as I still haven't figured out how to convert my dataset So, my question kinda reduces to: how to think about this coordinate The camera is at y=1.2, whereas it should be 1.0. The the camera seems to be at x=1, and z=0.8, whereas I specified x to beĠ.8, and z to be 1.2. The position of the camera relative to the unit cube seems wrong. The camera and the unit cube have different orientations for X, Y,Īnd Z (i.e.: red, green, blue axis). Notice how the camera I set up is having identity rotational part, and I'm trying to wrap my head around the coordinate system, and it looks On Fri, at 5:07 PM Martijn Courteaux wrote: If your point cloud data was collected in a local coordinate system, you can use an extrinsic matrix of the sensor used to collect the data to convert it to a. however the coordinate flipping is hard coded You can adjust the scale and offset components of the transformation byĪdding extra parameters to the json, see nerf_loader.cu that looks for keys In the end, I am afraid it is what it is. Over time, the 'original nerf' format of transforms.json became theĭominant (only practical) way to get data into ngp so, it comes across asĬonfusing as we don't explicitly talk about the mapping. Original nerf paper conventions to ngp's conventions'. So the snippet you found could be described as 'converting from the Have a different convention for 'up', camera up, etc. Origin at 0, scale the cameras to be around 3 units from the origin, and however, we wanted to beĬompatible specifically with the original nerf datasets, which place the that's just the convention we chose early on. ![]() Sorry for the shift in coordinate system! internally, ngp uses entirely 0-1īounding box, as displayed in the gui, with cameras looking down positive ![]() So, my question kinda reduces to: how to think about this coordinate system, as I still haven't figured out how to convert my dataset transformations to yours.īeta Was this translation helpful? Give feedback. Auto nerf_matrix_to_ngp( const Eigen::Matrix& nerf_matrix) * scale + offset
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |