Cross-Modal Distortion Approximation for Fast Bit Allocation of Video-Based Point Cloud Compression

In video-based point cloud compression (V-PCC), the optimal allocation of the total bitrate between geometry and color is a challenging but rewarding problem. Existing bit allocation approaches leverage statistical models to describe the rate and distortion of geometry and color information as functions of V-PCC quantization steps. However, to obtain the parameters of statistical models, these methods need to perform pre-coding for input point clouds for multiple times, resulting in high computational complexity. Consequently, the capability of these methods for practical application is limited. To address this problem, we derive the rate and distortion models based on projected images produced in the V-PCC encoding process, transforming the expensive point cloud pre-coding process into an efficient image pre-coding process. By utilizing the image-based distortion and rate models, the bit allocation problem is further formulated as a constrained convex optimization problem. Experimental results demonstrate that the proposed method exhibits significantly lower time complexity and higher rate-distortion performance compared to the existing methods.