Text this: DSC-MVSNet: attention aware cost volume regularization based on depthwise separable convolution for multi-view stereo