Learning Intrinsic Image Decomposition from Watching the World

TL;DR

This paper introduces a novel learning framework for intrinsic image decomposition that leverages temporal sequences of scenes under changing illumination, enabling training without ground truth data and improving generalization across diverse datasets.

Contribution

The paper proposes a new method that learns from image sequences over time, avoiding the need for ground truth decompositions and enhancing cross-dataset generalization.

Findings

The method generalizes well to multiple datasets.

It effectively learns from sequences without ground truth.

It achieves competitive performance on benchmark datasets.

Abstract

Single-view intrinsic image decomposition is a highly ill-posed problem, and so a promising approach is to learn from large amounts of data. However, it is difficult to collect ground truth training data at scale for intrinsic images. In this paper, we explore a different approach to learning intrinsic images: observing image sequences over time depicting the same scene under changing illumination, and learning single-view decompositions that are consistent with these changes. This approach allows us to learn without ground truth decompositions, and to instead exploit information available from multiple images when training. Our trained model can then be applied at test time to single views. We describe a new learning framework based on this idea, including new loss functions that can be efficiently evaluated over entire sequences. While prior learning-based methods achieve good performance on specific benchmarks, we show that our approach generalizes well to several diverse datasets, including MIT intrinsic images, Intrinsic Images in the Wild and Shading Annotations in the Wild.