I can try answer from the practical perspective, assuming you want to use projective reconstruction, if for example, you do not know the intrinsics of the cameras:
First, if the cameras are calibrated, you can project the feature from each image to a 3D ray and use SVD to find the 3D point in space. But, if the cameras are not calibrated, the SVD solution produces a point in some arbitrary projective coordinate system, which is not related to the true 3D positions. In this case, you cannot recover the actual 3D location from the SVD result without additional calibration information.