Direct Intrinsics: Learning Albedo-Shading Decomposition by Convolutional Regression

TL;DR

This paper presents a CNN-based method called direct intrinsics for decomposing images into albedo and shading, trained on synthetic data, outperforming prior methods on synthetic and generalizing well to real images.

Contribution

Introduces a CNN approach for intrinsic image decomposition trained on synthetic data, bypassing classical priors and inference algorithms.

Findings

Outperforms prior methods on Sintel dataset

Generalizes reasonably to real MIT dataset images

Uses only RGB input for decomposition

Abstract

We introduce a new approach to intrinsic image decomposition, the task of decomposing a single image into albedo and shading components. Our strategy, which we term direct intrinsics, is to learn a convolutional neural network (CNN) that directly predicts output albedo and shading channels from an input RGB image patch. Direct intrinsics is a departure from classical techniques for intrinsic image decomposition, which typically rely on physically-motivated priors and graph-based inference algorithms. The large-scale synthetic ground-truth of the MPI Sintel dataset plays a key role in training direct intrinsics. We demonstrate results on both the synthetic images of Sintel and the real images of the classic MIT intrinsic image dataset. On Sintel, direct intrinsics, using only RGB input, outperforms all prior work, including methods that rely on RGB+Depth input. Direct intrinsics also generalizes across modalities; it produces quite reasonable decompositions on the real images of the MIT dataset. Our results indicate that the marriage of CNNs with synthetic training data may be a powerful new technique for tackling classic problems in computer vision.