Fixes some implementation differences in the lstsq_grad method that cause less good performance of this algorithm. These differences include using the farplane gradient directly in the nearplane updates, adding some tiny constants to prevent instability, and only using real values for the optimal steps.
Fixes some implementation differences in the lstsq_grad method that cause less good performance of this algorithm. These differences include using the farplane gradient directly in the nearplane updates, adding some tiny constants to prevent instability, and only using real values for the optimal steps.