Where is Eq. (2) - (6) in the paper defined in the code?

[bchao1]

No description provided.

[niujinshuchong]

Hi, they are in the old version of code. You can find it here:

gaussian-opacity-fields/submodules/diff-gaussian-rasterization/cuda_rasterizer/forward.cu

Lines 504 to 517 in 5b21b8d

	// transform camera center and ray to gaussian's local coordinate system
	// current center is zero
	float3 cam_pos_local = {view2gaussian_j[12], view2gaussian_j[13], view2gaussian_j[14]};
	float3 ray_local = transformPoint4x3_without_t(ray_point, view2gaussian_j);
	// scale the ray_local and cam_pos_local
	double3 ray_local_scaled = { ray_local.x / scale_j.x, ray_local.y / scale_j.y, ray_local.z / scale_j.z };
	double3 cam_pos_local_scaled = { cam_pos_local.x / scale_j.x, cam_pos_local.y / scale_j.y, cam_pos_local.z / scale_j.z };
	// compute the minimal value
	// use AA, BB, CC so that the name is unique
	double AA = ray_local_scaled.x * ray_local_scaled.x + ray_local_scaled.y * ray_local_scaled.y + ray_local_scaled.z * ray_local_scaled.z;
	double BB = 2 * (ray_local_scaled.x * cam_pos_local_scaled.x + ray_local_scaled.y * cam_pos_local_scaled.y + ray_local_scaled.z * cam_pos_local_scaled.z);
	double CC = cam_pos_local_scaled.x * cam_pos_local_scaled.x + cam_pos_local_scaled.y * cam_pos_local_scaled.y + cam_pos_local_scaled.z * cam_pos_local_scaled.z;

.

But there are redundant computations so we merge it for efficiency. You can check here #58