Why does the FP16 model generate wavy noise on flat areas of the point cloud, while the FP32 model is smooth? Inference engine: TensorRT-10.14.1.48