rendering with custom cameras Matrix

[airchange]

trafficstars

When using nerf/instant-ngp before, each pixel of its image is actually represented by the starting point and end point of each ray（ray_o and rays_d） when rendering. It uses create_meshgrid() to generate a three-dimensional tensor and then operates to get ray_o and rays_d, so I can customize the ray to get every pixel I need. So in 3DGS, can its camera parameters(Matrix of ray_o and rays_d) be passed into the render pipeline in a similar way?