Learning Illuminant Estimation from Object Recognition
This addresses color constancy in computer vision, offering a novel training approach that avoids the need for labeled illuminant data, though it is incremental as it builds on existing deep learning methods.
The paper tackles illuminant estimation in images by training a deep learning model without ground truth illuminants, using object recognition as an auxiliary task, and shows it outperforms most deep learning methods in cross-dataset evaluations with competitive results against parametric solutions.
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.