About projection.

[cvg]

About projection.

[Zzh2000]

We follow the nerf-pytorch code convention and define the z-axis as "front to back".

[calmelo]

We follow the nerf-pytorch code convention and define the z-axis as "front to back".

Why not cam_cord[:, 2] *= -1 but cam_cord[:, 0] *= -1?

[calmelo]

We follow the nerf-pytorch code convention and define the z-axis as "front to back".

cam_cord[:, 0] *= -1是为了让计算得到的相机坐标的x, y, z在数据集的坐标系中均取反值吗？为什么不把cam_cord[:, 1] *= -1, cam_cord[:, 2] *= -1, 然后将计算得到的相机坐标转到数据集的坐标系中呢？
取反值之后, 因为u=x/z, v=y/z, 所以得到的uv最后就在数据集图片的坐标系内，然后z取反转到数据集图片坐标系中。
不知道我的理解正不正确，恳请大佬赐教！

[h8c2]

I am also confused about c2w[:3, 1] *= -1 c2w[:3, 2] *= -1 in load_poses in datasets.py. What does it mean to multiply a column of positions by negative one？

[h8c2]

我的理解是在代码相机（局部）坐标系下表达的坐标都是右上后的（正常相机坐标系为右下前，这里在load_pose就取反了），但是内参矩阵是右下前的，要变到像素坐标系就将第一列乘以负一，得到正常的像素坐标系取值范围