Latitude‐Guided Adaptive Inter‐Frame CU Partitioning for Fast Panoramic Video Encoding
Wenjun Zheng, Chao Yang, Ping An, Xinpeng Huang · Electronics Letters · 2025
ABSTRACT This letter proposes a multi‐feature attentive convolutional neural network (MFA‐CNN)‐based fast coding unit (CU) partitioning algorithm for panoramic video inter‐coding in versatile video coding (VVC). This framework incorporates a U‐Net architecture embedded with channel attention mechanisms to fuse multi‐dimensional features, enabling accurate prediction of CU partitioning paths and early termination of low‐probability partition modes. To address coding efficiency in panoramic videos, the algorithm employs latitude‐guided adaptive partitioning threshold and dynamic quantisation parameter (QP) adjustment strategies: Finer partitioning is applied to equatorial regions with complex textures, while redundant partitioning is terminated early in polar regions dominated by homogeneous textures. Experimental validation shows 33.64% average encoding time savings with merely 0.93% rate‐distortion (RD) performance degradation, outperforming existing state‐of‐the‐art methods.