Sill-Net: Feature Augmentation with Separated Illumination Representation
arXiv:2102.03539
Abstract
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.
References in corpus (20)
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
- Prototypical Networks for Few-shot Learning
- mixup: Beyond Empirical Risk Minimization
- Instance Normalization: The Missing Ingredient for Fast Stylization
- Wide Residual Networks
- A Simple Neural Attentive Meta-Learner
- Data Augmentation Generative Adversarial Networks
- A Closer Look at Memorization in Deep Networks
- GAN Augmentation: Augmenting Training Data using Generative Adversarial Networks
- Data Augmentation by Pairing Samples for Images Classification
- Disentangling factors of variation in deep representations using adversarial training
- Multi-Level Variational Autoencoder: Learning Disentangled Representations from Grouped Observations
- Leveraging the Feature Distribution in Transfer-based Few-Shot Learning
- Meta-Learned Confidence for Few-shot Learning
- Regularizing Deep Networks with Semantic Data Augmentation
- Prototype Rectification for Few-Shot Learning
- Graph-based Interpolation of Feature Vectors for Accurate Few-Shot Classification
- Combining Self-Supervised and Supervised Learning with Noisy Labels
- Adversarial Disentanglement with Grouped Observations