ScholarWorks@성균관대학교

초록

Neural network-based video coding tools have been actively explored to improve video coding efficiency. However, the high computational complexity and dependence on floating-point operations present significant to real-world deployment. In this work, we investigate a hardware-friendly neural network-based chroma intra prediction (HF-NNCIP) with luma attention integrated into the Versatile Video Coding (VVC) framework. To achieve its hardware-friendly design, we carry out architectural simplification, network quantization, and selective inference. Experimental evaluations demonstrate that the proposed method achieves a significant acceleration in processing, with encoding and decoding speeds improved by approximately 3.9 × and 7.8 ×, respectively, compared to conventional approaches with only a marginal compromise in coding efficiency. The proposed approach suggests a practical direction for designing complex neural network-based tools in a hardware-friendly manner, enabling their potential integration into real-world hardware-constrained environments.

키워드