Sill-Net: Feature Augmentation with Separated Illumination Representation
This work aims to improve the robustness of deep neural networks for visual object recognition under varying illumination conditions, which is a common problem for computer vision practitioners.
This paper addresses the challenge of illumination variations in visual object recognition by proposing Sill-Net, a novel neural network architecture. Sill-Net separates illumination features from images and uses them to augment training samples in the feature space, leading to superior performance on several object classification benchmarks.
For visual object recognition tasks, the illumination variations can cause distinct changes in object appearance and thus confuse the deep neural network based recognition models. Especially for some rare illumination conditions, collecting sufficient training samples could be time-consuming and expensive. To solve this problem, in this paper we propose a novel neural network architecture called Separating-Illumination Network (Sill-Net). Sill-Net learns to separate illumination features from images, and then during training we augment training samples with these separated illumination features in the feature space. Experimental results demonstrate that our approach outperforms current state-of-the-art methods in several object classification benchmarks.