Improving Semantic Segmentation via Self-Training

Abstract

Deep learning usually achieves the best results with complete supervision. In the case of semantic segmentation, this means that large amounts of pixelwise annotations are required to learn accurate models. In this paper, we show that we can obtain state-of-the-art results using a semi-supervised approach, specifically a self-training paradigm. We first train a teacher model on labeled data, and then generate pseudo labels on a large set of unlabeled data. Our robust training framework can digest human-annotated and pseudo labels jointly and achieve top performances on Cityscapes, CamVid and KITTI datasets while requiring significantly less supervision. We also demonstrate the effectiveness of self-training on a challenging cross-domain generalization task, outperforming conventional finetuning method by a large margin. Lastly, to alleviate the computational burden caused by the large amount of pseudo labels, we propose a fast training schedule to accelerate the training of segmentation models by up to 2x without performance degradation.

References in corpus (19)

• ADVENT: Adversarial Entropy Minimization for Domain Adaptation in Semantic Segmentation
• ParseNet: Looking Wider to See Better
• Momentum Contrast for Unsupervised Visual Representation Learning
• High-Resolution Representations for Labeling Pixels and Regions
• Fast-SCNN: Fast Semantic Segmentation Network
• Searching for Efficient Multi-Scale Architectures for Dense Image Prediction
• Billion-scale semi-supervised learning for image classification
• Self-training with Noisy Student improves ImageNet classification
• Wider or Deeper: Revisiting the ResNet Model for Visual Recognition
• FastFCN: Rethinking Dilated Convolution in the Backbone for Semantic Segmentation
• Category Anchor-Guided Unsupervised Domain Adaptation for Semantic Segmentation
• High-performance Semantic Segmentation Using Very Deep Fully Convolutional Networks
• Semi and Weakly Supervised Semantic Segmentation Using Generative Adversarial Network
• Domain Adaptation for Semantic Segmentation via Class-Balanced Self-Training
• Constructing Self-motivated Pyramid Curriculums for Cross-Domain Semantic Segmentation: A Non-Adversarial Approach
• SPGNet: Semantic Prediction Guidance for Scene Parsing
• Pixel-wise Attentional Gating for Parsimonious Pixel Labeling
• Structured Knowledge Distillation for Dense Prediction
• Label-Driven Reconstruction for Domain Adaptation in Semantic Segmentation

Cited by in corpus (9)

• PseudoSeg: Designing Pseudo Labels for Semantic Segmentation
• Self-paced and self-consistent co-training for semi-supervised image segmentation
• Semi-Supervised Semantic Segmentation with Cross Pseudo Supervision
• Naive-Student: Leveraging Semi-Supervised Learning in Video Sequences for Urban Scene Segmentation
• MetaCorrection: Domain-aware Meta Loss Correction for Unsupervised Domain Adaptation in Semantic Segmentation
• Re-distributing Biased Pseudo Labels for Semi-supervised Semantic Segmentation: A Baseline Investigation
• A Unified Efficient Pyramid Transformer for Semantic Segmentation
• AutoAdapt: Automated Segmentation Network Search for Unsupervised Domain Adaptation
• Multi-Object Tracking with Hallucinated and Unlabeled Videos