ReasonX：多模态大语言模型引导的本征图像分解
ReasonX: MLLM-Guided Intrinsic Image Decomposition
把多模态大语言模型当作感知评判器，用其输出的相对本征比较构造 GRPO 奖励，从而在无标注真实图像上微调本征分解模型；这一「MLLM 比较式监督」范式可迁移到深度、法线、反照率等任何可定义成对关系的稠密预测任务上。
原文依据 uses these comparisons as GRPO rewards for fine-tuning intrinsic decomposition models on unlabeled, in-the-wild images
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论本征图像分解旨在将图像分离为反照率、深度、法线与光照等物理成分。尽管近期基于扩散模型与 Transformer 的模型受益于合成数据集提供的成对监督，但其在多样化的真实场景中的泛化仍具挑战。我们提出 ReasonX，一种新颖框架，它利用多模态大语言模型（MLLM）作为感知评判器提供相对的本征比较，并将这些比较作为 GRPO 奖励，在无标注的真实图像上微调本征分解模型。与面向生成模型的强化学习方法不同，我们的框架通过奖励评判器的关系性评估与由模型输出解析推导出的关系之间的一致性，来对齐条件本征预测器。ReasonX 与模型无关，可应用于不同的本征预测器。在多种基础架构与模态上，ReasonX 均带来显著提升，包括在 IIW 反照率上实现 9–25% 的 WHDR 降低，以及在 ETH3D 上最高 46% 的深度精度提升，凸显了 MLLM 引导的比较式监督在连接低层与高层视觉推理方面的潜力。
英文原文摘要 1191 字符
Intrinsic image decomposition aims to separate images into physical components such as albedo, depth, normals, and illumination. While recent diffusion- and transformer-based models benefit from paired supervision from synthetic datasets, their generalization to diverse, real-world scenarios remains challenging. We propose ReasonX, a novel framework that leverages a multimodal large language model (MLLM) as a perceptual judge providing relative intrinsic comparisons, and uses these comparisons as GRPO rewards for fine-tuning intrinsic decomposition models on unlabeled, in-the-wild images. Unlike RL methods for generative models, our framework aligns conditional intrinsic predictors by rewarding agreement between the judge's relational assessments and analytically derived relations from the model's outputs. ReasonX is model-agnostic and can be applied to different intrinsic predictors. Across multiple base architectures and modalities, ReasonX yields significant improvements, including 9-25% WHDR reduction on IIW albedo and up to 46% depth accuracy gains on ETH3D, highlighting the promise of MLLM-guided comparative supervision to bridge low- and high-level vision reasoning.