超越 8 比特：HDR-UGC 视频的主客观质量评估
Seeing Beyond 8bits: Subjective and Objective Quality Assessment of HDR-UGC Videos
提供了 HDR-UGC 主观质量数据集与 HDR 感知视觉编码器，可复用于任何需要高动态范围输入的感知模型；HAPO 的 HDR-SDR 对比 KL 与高斯加权回归奖励，是把 RL 微调锚定到特定成像线索上的可迁移范式。
原文依据 we curate HDR-UGC-44K, a large-scale subjective dataset of ~44K videos from 6.5K sources with >1.5M crowd ratings
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论高动态范围（HDR）用户生成（UGC）视频正在社交平台上迅速增长，然而大多数感知视频质量评估（VQA）系统仍针对标准动态范围（SDR）设计。HDR 更高的位深、更宽的色域和更高的亮度范围会暴露出近黑压暗、高光裁剪、色带和曝光闪烁等失真，这些失真放大了 UGC 伪影并对 SDR 模型构成挑战。为促进该方向的发展，我们构建了 HDR-UGC-44K，这是一个大规模主观数据集，包含来自 6.5K 个源视频的约 44K 条视频以及超过 150 万条众包评分，覆盖多样场景、拍摄条件和压缩设置。我们进一步提出 HDR-Q，这是首个面向 HDR-UGC 视频质量评估的多模态大语言模型（MLLM）。我们提出：(i) 一种新颖的 HDR 感知视觉编码器，用于产生对 HDR 敏感的嵌入；(ii) HDR 感知策略优化（HAPO），一个将推理锚定到 HDR 线索上的强化学习微调框架。HAPO 通过 HDR-SDR 对比 KL 项增强 GRPO，鼓励词元依赖 HDR 输入，并采用高斯加权回归奖励实现细粒度的 MOS 校准。在 HDR-UGC-44K 和公开 HDR-VQA 基准上，HDR-Q 均取得最优性能。项目主页：https://shreshthsaini.github.io/Beynod8Bits
英文原文摘要 1253 字符
High Dynamic Range (HDR) user-generated (UGC) videos are rapidly proliferating across social platforms, yet most perceptual video quality assessment (VQA) systems remain tailored to Standard Dynamic Range (SDR). HDR's higher bit depth, wide color gamut, and elevated luminance range expose distortions such as near-black crushing, highlight clipping, banding, and exposure flicker that amplify UGC artifacts and challenge SDR models. To catalyze progress, we curate HDR-UGC-44K, a large-scale subjective dataset of ~44K videos from 6.5K sources with >1.5M crowd ratings, spanning diverse scenes, capture conditions, and compression settings. We further introduce HDR-Q, the first Multimodal Large Language Model (MLLM) for HDR-UGC VQA. We propose (i) a novel HDR-aware vision encoder to produce HDR-sensitive embeddings, and (ii) HDR-Aware Policy Optimization (HAPO), an RL finetuning framework that anchors reasoning to HDR cues. HAPO augments GRPO via an HDR-SDR contrastive KL that encourages token reliance on HDR inputs and a gaussian weighted regression reward for fine-grained MOS calibration. Across HDR-UGC-44K and public HDR-VQA benchmarks, HDR-Q delivers state-of-the-art performance. Project page: https://shreshthsaini.github.io/Beynod8Bits