Dense Relation Network: Learning Consistent and Context-Aware Representation for Semantic Image Segmentation

By Yueqing Zhuang, Fan Yang, Li Tao, Cong Ma, Ziwei Zhang, Yuan Li, Huizhu Jia, Xiaodong Xie, Wen Gao. 2018

42 citations recorded by the source corpus.

Semantic image segmentation, which aims at assigning pixel-wise category, is one of challenging image understanding problems. Global context plays an important role on local pixel-wise category assignment. To make the best of global context, in this paper, we propose dense relation network (DRN) and context-restricted loss (CRL) to aggregate global and local information. DRN uses Recurrent Neural Network (RNN) with different skip lengths in spatial directions to get context-aware representations while CRL helps aggregate them to learn consistency. Compared with previous methods, our proposed method takes full advantage of hierarchical contextual representations to produce high-quality results. Extensive experiments demonstrate that our method achieves significant state-of-the-art performances on Cityscapes and Pascal Context benchmarks, with mean-IoU of 82.8% and 49.0% respectively.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Andreas Bär
2. Marvin Klingner
3. Tim Fingscheidt
4. Philipp Donn

Citations in the corpus

• Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Image Segmentation (2019, neutral): “FRRN-A [60] 63.0 GridNet [17] 69.5 FRRN-B [60] 71.8 Auto-DeepLab-S 79.9 Auto-DeepLab-L 80.4 Auto-DeepLab-S 80.9 Auto-DeepLab-L 82.1 ResNet-38 [82] 80.6 PSPNet [88] 81.2 Mapillary [4] 82.0 DeepLabv3+ [11] 82.1 DPC [6] 82.7 DRN CRL Coarse [91] 82.8 In Tab.”
• Improving Semantic Segmentation via Video Propagation and Label Relaxation (2018, unknown): “1 DRN-CRL [46] test 98.”
• Structured Consistency Loss for semi-supervised semantic segmentation (2020, unknown): “More recently, Chen et al. [5] suggestes the well-organized architecture with combining encoder-decoder architecture and dilated convolution, which have been followed by many subsequent methods to achieve state-of-the-art performance [12, 21, 30].”
• Class-Incremental Learning for Semantic Segmentation Re-Using Neither Old Data Nor Old Labels (2020, methodology): “Henceforth, major improvements in network architecture and training procedure [2], [3], [4], [5], [6], [7] boosted the performance of such models without changing the basic concept.”