Sill-Net: Feature Augmentation with Separated Illumination Representation

6 Feb 2021  ·  Haipeng Zhang, Zhong Cao, Ziang Yan, ChangShui Zhang ·

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.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Traffic Sign Recognition BelgaLogos Sill-Net Accuracy 89.48 # 1
Traffic Sign Recognition Belgian Traffic Sign Classification Sill-Net Accuracy 98.97 # 1
Traffic Sign Recognition Chinese Traffic Sign Database Sill-Net Accuracy 97.19 # 1
Few-Shot Image Classification CIFAR-FS 5-way (1-shot) Illumination Augmentation Accuracy 87.73 # 4
Few-Shot Image Classification CIFAR-FS 5-way (5-shot) Illumination Augmentation Accuracy 91.09 # 4
Few-Shot Image Classification CUB 200 5-way 1-shot Illumination Augmentation Accuracy 94.73 # 4
Few-Shot Image Classification CUB 200 5-way 5-shot Illumination Augmentation Accuracy 96.28 # 3
Traffic Sign Recognition FlickrLogos-32 Sill-Net Accuracy 95.80 # 1
Traffic Sign Recognition GTSRB Sill-Net Accuracy 99.68% # 2
Few-Shot Image Classification Mini-Imagenet 5-way (1-shot) Illumination Augmentation Accuracy 82.99 # 8
Few-Shot Image Classification Mini-Imagenet 5-way (5-shot) Illumination Augmentation Accuracy 89.14 # 10
Traffic Sign Recognition TopLogo-10 Sill-Net Accuracy 89.66 # 1
Traffic Sign Recognition Tsinghua-Tencent 100K Sill-Net Accuracy 99.53 # 1

Methods


No methods listed for this paper. Add relevant methods here