Self-Supervised Intrinsic Image Decomposition

TL;DR

This paper introduces RIN, a self-supervised model for intrinsic image decomposition that leverages unsupervised reconstruction to improve predictions of reflectance, shape, and lighting from a single image, enabling effective learning from unlabeled data.

Contribution

The paper presents a novel self-supervised approach that combines an image decomposition pipeline with a learned shading model, allowing training on unlabeled data and transfer to unseen categories.

Findings

Performs well on intrinsic image decomposition tasks.

Enables transfer learning to unseen object categories and lighting conditions.

Utilizes large-scale unlabeled data effectively during training.

Abstract

Intrinsic decomposition from a single image is a highly challenging task, due to its inherent ambiguity and the scarcity of training data. In contrast to traditional fully supervised learning approaches, in this paper we propose learning intrinsic image decomposition by explaining the input image. Our model, the Rendered Intrinsics Network (RIN), joins together an image decomposition pipeline, which predicts reflectance, shape, and lighting conditions given a single image, with a recombination function, a learned shading model used to recompose the original input based off of intrinsic image predictions. Our network can then use unsupervised reconstruction error as an additional signal to improve its intermediate representations. This allows large-scale unlabeled data to be useful during training, and also enables transferring learned knowledge to images of unseen object categories, lighting conditions, and shapes. Extensive experiments demonstrate that our method performs well on both intrinsic image decomposition and knowledge transfer.