ReasonX: MLLM-Guided Intrinsic Image Decomposition

TL;DR

ReasonX introduces a novel framework that uses multimodal large language models as perceptual judges to improve intrinsic image decomposition, especially on real-world images, by leveraging relational comparisons as rewards for fine-tuning models.

Contribution

It proposes a model-agnostic, MLLM-guided supervision method that enhances intrinsic decomposition models without requiring labeled data, bridging the gap between synthetic training and real-world application.

Findings

9-25% WHDR reduction on IIW albedo

Up to 46% depth accuracy gains on ETH3D

Effective across multiple architectures and modalities

Abstract

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.