Text this: Learnable Resized and Laplacian-Filtered U-Net: Better Road Marking Extraction and Classification on Sparse-Point-Cloud-Derived Imagery