ADVENT: Adversarial Entropy Minimization for Domain Adaptation in Semantic Segmentation

Vu, Tuan-Hung; Jain, Himalaya; Bucher, Maxime; Cord, Matthieu; Pérez, Patrick

doi:10.1109/cvpr.2019.00262

Cited by 1,157 publications

(1,159 citation statements)

References 37 publications

Supporting

Mentioning

1,063

Contrasting

Order By: Relevance

“…Recently, inspired by semi-supervised learning [11,17] which also utilizes the unlabeled data, there are several semisupervised learning based methods [9,24,36,31] proposed for UDA task. Assuming that areas with higher prediction probability are more accurate, the class-balanced selftraining [36] generated pseudo labels based on class-wise thresholds.…”

Section: Related Workmentioning

confidence: 99%

“…Thus making unlabeled samples less ambiguous can help classes to be more separable, e.g., minimizing the conditional entropy [11]. ADVENT [31] adopted this idea in the UDA field and minimized the prediction entropy of the target sample.…”

Section: Related Workmentioning

confidence: 99%

“…Entropy Minimization. In the [31], they try to minimize the Shannon entropy of the target sample prediction. Thus, their objective function for target samples is:…”

Section: Overview Of Udamentioning

confidence: 99%

“…Entropy minimization [11], which encourages unambiguous cluster assignments, is one of the most popular methods in semi-supervised learning. ADVENT [31] directly adopts the entropy minimization method to UDA for semantic segmentation, but their result is inferior to state-of-the-art approaches.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Domain Adaptation for Semantic Segmentation With Maximum Squares Loss

Chen

Xue

Cai

2019

2019 IEEE/CVF International Conference on Computer Vision (ICCV)

279

184

View full text Add to dashboard Cite

Deep neural networks for semantic segmentation always require a large number of samples with pixel-level labels, which becomes the major difficulty in their real-world applications. To reduce the labeling cost, unsupervised domain adaptation (UDA) approaches are proposed to transfer knowledge from labeled synthesized datasets to unlabeled real-world datasets. Recently, some semi-supervised learning methods have been applied to UDA and achieved state-of-the-art performance. One of the most popular approaches in semi-supervised learning is the entropy minimization method. However, when applying the entropy minimization to UDA for semantic segmentation, the gradient of the entropy is biased towards samples that are easy to transfer. To balance the gradient of well-classified target samples, we propose the maximum squares loss. Our maximum squares loss prevents the training process being dominated by easy-to-transfer samples in the target domain. Besides, we introduce the image-wise weighting ratio to alleviate the class imbalance in the unlabeled target domain. Both synthetic-to-real and cross-city adaptation experiments demonstrate the effectiveness of our proposed approach. The code is released at https://github. com/ZJULearning/MaxSquareLoss.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

“…Entropy Minimization. In the [31], they try to minimize the Shannon entropy of the target sample prediction. Thus, their objective function for target samples is:…”

Section: Overview Of Udamentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Domain Adaptation for Semantic Segmentation With Maximum Squares Loss

Chen

Xue

Cai

2019

2019 IEEE/CVF International Conference on Computer Vision (ICCV)

279

184

View full text Add to dashboard Cite

show abstract

“…Feature alignment of the two domains in the latent feature space is further developed in [32] through an information bottleneck before the adversarial adaptation module on the feature space. Following a self-training strategy, Vu et al [29] resort to an entropy-based loss. They explore both a direct approach with a hand-crafted loss and indirect adversarial solution for which the objective is expressed by a learnable discriminator.…”

Section: Related Workmentioning

confidence: 99%

Unsupervised domain adaptation for mobile semantic segmentation based on cycle consistency and feature alignment

Toldo

Michieli

Agresti

et al. 2020

Image and Vision Computing

View full text Add to dashboard Cite

The supervised training of deep networks for semantic segmentation requires a huge amount of labeled real world data. To solve this issue, a commonly exploited workaround is to use synthetic data for training, but deep networks show a critical performance drop when analyzing data with slightly different statistical properties with respect to the training set. In this work, we propose a novel Unsupervised Domain Adaptation (UDA) strategy to address the domain shift issue between real world and synthetic representations. An adversarial model, based on the cycle consistency framework, performs the mapping between the synthetic and real domain. The data is then fed to a MobileNet-v2 architecture that performs the semantic segmentation task. An additional couple of discriminators, working at the feature level of the MobileNet-v2, allows to better align the features of the two domain distributions and to further improve the performance. Finally, the consistency of the semantic maps is exploited. After an initial supervised training on synthetic data, the whole UDA architecture is trained end-to-end considering all its components at once. Experimental results show how the proposed strategy is able to obtain impressive performance in adapting a segmentation network trained on synthetic data to real world scenarios. The usage of the lightweight MobileNet-v2 architecture allows its deployment on devices with limited computational resources as the ones employed in autonomous vehicles.

show abstract

Fusing feature and output space for unsupervised domain adaptation on medical image segmentation

Wang

et al. 2023

Int J Imaging Syst Tech

View full text Add to dashboard Cite

Image segmentation requires large amounts of annotated data. However, collecting massive datasets with annotations is difficult since they are expensive and labor‐intensive. The unsupervised domain adaptation (UDA) for image segmentation is a promising approach to address the label‐scare problem on the target domain, which enables the trained model on the source labeled domain to be adaptive to the target domain. The adversarial‐based methods encourage extracting the domain‐invariant features by training a domain discriminator to mitigate the domain gap. Existing UDA segmentation methods fail to obtain satisfied segmentation results as they only consider the global knowledge of output space while neglecting the local information of feature space. In this paper, a fusing feature and output (FFO) space method is proposed for UDA, which in the context of medical image segmentation. The proposed model is learned by training a more powerful domain discriminator, which considers features extracted from both feature space and output space. Extensive experiments carried out on several medical image datasets show the adaptation effectiveness of our approach in improving the segmentation performance.

show abstract

ADVENT: Adversarial Entropy Minimization for Domain Adaptation in Semantic Segmentation

Cited by 1,157 publications

References 37 publications

Domain Adaptation for Semantic Segmentation With Maximum Squares Loss

Domain Adaptation for Semantic Segmentation With Maximum Squares Loss

Unsupervised domain adaptation for mobile semantic segmentation based on cycle consistency and feature alignment

Fusing feature and output space for unsupervised domain adaptation on medical image segmentation

Contact Info

Product

Resources

About