GT-SVJ：基于生成Transformer的自监督视频裁判用于高效视频奖励建模
GT-SVJ: Generative-Transformer-Based Self-Supervised Video Judge For Efficient Video Reward Modeling
将视频生成模型（如扩散模型）通过能量模型重参数化，并利用潜在空间扰动构造负样本进行对比训练，从而作为时序感知的奖励模型，为视频生成质量评估和偏好对齐提供了新思路。
原文依据 repurposing video generative models, which are inherently designed to model temporal structure, as reward models
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论将视频生成模型与人类偏好对齐仍然具有挑战性：当前方法依赖视觉语言模型（VLM）进行奖励建模，但这些模型难以捕捉微妙的时序动态。我们提出了一种根本不同的方法：重新利用本质上设计用于建模时序结构的视频生成模型作为奖励模型。我们提出了基于生成Transformer的自监督视频裁判（GT-SVJ），这是一种新颖的评估模型，将最优的视频生成模型转化为强大的时序感知奖励模型。我们的关键见解是，生成模型可以被重新表述为能量模型（EBM），该模型为高质量视频分配低能量，为退化视频分配高能量，从而在通过对比目标训练时能够以显著的精度区分视频质量。为了防止模型利用真实视频和生成视频之间的表面差异，我们通过受控的潜在空间扰动设计了具有挑战性的合成负视频：时序切片、特征交换和帧混洗，这些扰动模拟了真实但微妙的视觉退化。这迫使模型学习有意义的时空特征，而不是琐碎的伪影。GT-SVJ仅使用3万个人工标注就在GenAI-Bench和MonteBench上达到了最优性能：比现有的基于VLM的方法少6倍到65倍。
英文原文摘要 1440 字符
Aligning video generative models with human preferences remains challenging: current approaches rely on Vision-Language Models (VLMs) for reward modeling, but these models struggle to capture subtle temporal dynamics. We propose a fundamentally different approach: repurposing video generative models, which are inherently designed to model temporal structure, as reward models. We present the Generative-Transformer-based Self-Supervised Video Judge (\modelname), a novel evaluation model that transforms state-of-the-art video generation models into powerful temporally-aware reward models. Our key insight is that generative models can be reformulated as energy-based models (EBMs) that assign low energy to high-quality videos and high energy to degraded ones, enabling them to discriminate video quality with remarkable precision when trained via contrastive objectives. To prevent the model from exploiting superficial differences between real and generated videos, we design challenging synthetic negative videos through controlled latent-space perturbations: temporal slicing, feature swapping, and frame shuffling, which simulate realistic but subtle visual degradations. This forces the model to learn meaningful spatiotemporal features rather than trivial artifacts. \modelname achieves state-of-the-art performance on GenAI-Bench and MonteBench using only 30K human-annotations: 6xto 65xfewer than existing VLM-based approaches.