Conference
Open access
Sep 2026
Beyond Pipeline Mimicry: Expert-level Aesthetic ISP via Reward-Guided Flow
This work proposes AesISP, the first expert-level aesthetic ISP framework formulated as a reward flow model, and establishes a Privileged Prior Distillation paradigm, and introduces a multi-dimensional reward-driven reinforcement learning approach.
Tong Qiao, Kepeng Xu, Gang He et al.
· Proceedings of the Thirty-Fi... · 0 citations