PointNetV3: feature extraction with position encoding