Projection transformation calculation is a bit weird

[xubin1994]

f2-nerf/src/PtsSampler/PersSampler.cpp

Line 555 in 98f0daa

float focal = (intri.index({ 0, 0 }) / intri.index({ 0, 2 })).item<float>();

Hello author, I am analyzing the source code implementation. I have some questions about the calculation of frame_trans here. Generally speaking, we use a 44 external parameter matrix and a 33 internal parameter matrix to get the projected image coordinates. The calculation here Ways seem to merge in some way? Are there any references? I'm quite confused about focal length here, theoretically focal = (intri.index({ 0, 0 })