Thank you for sharing your reproduction work. I read your cuda code. I found that your calculations on AABB and center point are different from the author's. Regarding the calculation of AABB, the author has shared his calculation formula, which seems to be a more accurate calculation method. As for the calculation of center point, I initially thought the same as you, that is, to directly use the projection point of the surfel center, but after reading the author's AABB calculation method, I think it is more appropriate to use the AABB center than the surfel projection center, because when the surfel is relatively large, due to perspective projection, the projection point of the surfel may deviate more from the AABB center.
Thank you for sharing your reproduction work. I read your cuda code. I found that your calculations on AABB and center point are different from the author's. Regarding the calculation of AABB, the author has shared his calculation formula, which seems to be a more accurate calculation method. As for the calculation of center point, I initially thought the same as you, that is, to directly use the projection point of the surfel center, but after reading the author's AABB calculation method, I think it is more appropriate to use the AABB center than the surfel projection center, because when the surfel is relatively large, due to perspective projection, the projection point of the surfel may deviate more from the AABB center.
https://github.com/hbb1/diff-surfel-rasterization/issues/8#issuecomment-2138069016