Instance Adaptive Self-Training for Unsupervised Domain Adaptation

By Ke Mei, Chuang Zhu, Jiaqi Zou, Shanghang Zhang. European Conference on Computer Vision, 2020

349 citations recorded by the source corpus.

The divergence between labeled training data and unlabeled testing data is a significant challenge for recent deep learning models. Unsupervised domain adaptation (UDA) attempts to solve such a problem. Recent works show that self-training is a powerful approach to UDA. However, existing methods have difficulty in balancing scalability and performance. In this paper, we propose an instance adaptive self-training framework for UDA on the task of semantic segmentation. To effectively improve the quality of pseudo-labels, we develop a novel pseudo-label generation strategy with an instance adaptive selector. Besides, we propose the region-guided regularization to smooth the pseudo-label region and sharpen the non-pseudo-label region. Our method is so concise and efficient that it is easy to be generalized to other unsupervised domain adaptation methods. Experiments on 'GTA5 to Cityscapes' and 'SYNTHIA to Cityscapes' demonstrate the superior performance of our approach compared with the state-of-the-art methods.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

View Paper: https://www.semanticscholar.org/paper/3c581408468a3cd6c5ac2d852bae3a6a70fd2491

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

MetaCorrection: Domain-aware Meta Loss Correction for Unsupervised Domain Adaptation in Semantic Segmentation (2021, positive): “The general idea of UDA semantic segmentation methods is to perform domain alignment through adversarial learning [2, 6, 7, 9, 11, 20, 22, 26, 40, 41] or utilize self-training strategy on target samples to adapt the segmentation models [4, 17, 23, 28, 30, 35, 46, 48, 49, 50, 51].”
BAPA-Net: Boundary Adaptation and Prototype Alignment for Cross-domain Semantic Segmentation (2021, methodology): “The previous cross-domain semantic segmentation methods are mainly divided into three categories: 1) methods based on adversarial training including AdvEnt [52], BDL [30] and FADA [54], 2) methods based on selftraining including PyCDA [31], CRST [66], R-MRNet [64], UDADT [57] and IAST [41], and 3) m”
DACS: Domain Adaptation via Cross-domain Mixed Sampling (2020, background): “Variants include specialised sampling [60, 27] and handling of uncertainty [61, 56].”
LoveDA: A Remote Sensing Land-Cover Dataset for Domain Adaptive Semantic Segmentation (2021, methodology): “Mei et al. [25] used an instance adaptive self-training (IAST) selector for sample balance.”
MIC: Masked Image Consistency for Context-Enhanced Domain Adaptation (2022, methodology): “The target loss L is defined according to the UDA strategy such as adversarial training [8,21,69,86,87,91] or selftraining [34, 63, 85, 106, 107, 114].”
HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation (2022, neutral): “In order to regularize the training and to avoid pseudo-label drift, approaches such as confidence thresholding [48,99], pseudo-label prototypes [41,87,88], and consistency regularization [55, 57, 62] based on data augmentation [1, 10, 49, 51], different context [36,96], domain-mixup…”
Undoing the Damage of Label Shift for Cross-domain Semantic Segmentation (2022, methodology): “Combining with self-training Since the self-training methods [32, 44, 49] generally use domain alignment methods as their warm-up model to produce pseudo labels, our model can be easily plugged into them.”
DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation (2021, methodology): “Also in UDA, re-weighting [53, 99] and class-balanced sampling for image classification [58] were applied.”
OpenEarthMap: A Benchmark Dataset for Global High-Resolution Land Cover Mapping (2022, methodology): “IAST and CBST performance improves for water but they lose the ability to recognize the roads.”