Can LLMs Design Video Coding Tools? A Case Study on Planar Mode
An LLM-generated Planar predictor beat the conventional mode in VVenC’s faster preset, but only by a narrow margin.
The paper reports 0.18% bitrate savings with 0.4% complexity overhead on the standard benchmark. Its workflow has the model generate Planar predictors, test them in encoder trials, then revise from performance feedback. In ECM, the authors tested replacement and added-mode integration for directional Planar modes, with gains only under a constrained low-resolution setting. The authors frame the result as preliminary evidence, not a solved path for LLM-designed video coding tools. ArXiv · AI/CL/LG's note
The paper reports 0.18% bitrate savings with 0.4% complexity overhead on the standard benchmark. Its workflow has the model generate Planar predictors, test them in encoder trials, then revise from performance feedback. In ECM, the authors tested replacement and added-mode integration for directional Planar modes, with gains only under a constrained low-resolution setting. The authors frame the result as preliminary evidence, not a solved path for LLM-designed video coding tools. ArXiv · AI/CL/LG's note
score 4