Learning Illuminant Estimation from Object Recognition

TL;DR

This paper introduces a novel deep learning approach for illuminant estimation that is trained without ground truth illuminant labels, instead optimizing for auxiliary tasks like object recognition, and demonstrates competitive performance on standard datasets.

Contribution

First deep learning model for illuminant estimation trained without explicit illuminant annotations, leveraging auxiliary tasks to improve color constancy performance.

Findings

Outperforms most deep learning methods in cross-dataset evaluations

Achieves competitive results compared to parametric solutions

Demonstrates effectiveness of auxiliary task training for illuminant estimation

Abstract

In this paper we present a deep learning method to estimate the illuminant of an image. Our model is not trained with illuminant annotations, but with the objective of improving performance on an auxiliary task such as object recognition. To the best of our knowledge, this is the first example of a deep learning architecture for illuminant estimation that is trained without ground truth illuminants. We evaluate our solution on standard datasets for color constancy, and compare it with state of the art methods. Our proposal is shown to outperform most deep learning methods in a cross-dataset evaluation setup, and to present competitive results in a comparison with parametric solutions.