Pcl Integral Image Estimation throwing NAN's for normal fields

Question

When using pcl::IntegralImageEstimation to compute normals, the normal fields are populated with NAN's but when I use other estimation methods like NormalEstimationOMP I can generate normals. I don't get any errors as such, I want to use Integral method as it is much faster than the OMP version and can compute normals real-time for my use case. I have also tried other estimation methods in IntegralImageNormalEstimation (COVARIANCE_MATRIX, AVERAGE_3D_GRADIENT, AVERAGE_DEPTH_CHANGE, SIMPLE_3D_GRADIENT)

Here's the code I'm using;

        pcl::IntegralImageNormalEstimation<PointType, PointType> ne;
        ne.setNormalEstimationMethod (ne.AVERAGE_DEPTH_CHANGE);
        ne.setMaxDepthChangeFactor(2.0f);
        ne.setNormalSmoothingSize(10.0f);
        ne.setViewPoint(pose_in.pose.position.x, pose_in.pose.position.y, pose_in.pose.position.z);
        ne.setInputCloud(transform_cloud);
        SWRI_PROFILE("compute_normal");
        ne.compute(*cloud_out);