DSP: Dual Soft-Paste for Unsupervised Domain Adaptive Semantic Segmentation

Gao, Li; Zhang, Jing; Zhang, Lefei; Tao, Dacheng

doi:10.1145/3474085.3475186

Cited by 55 publications

(25 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [23], authors introduced the STransFuse model as a new semantic segmentation method for remote sensing images. Gao et al [24] proposed a novel unsupervised domain adaptive semantic segmentation method by selecting some classes from a source domain image and softly pasting the corresponding image patch on both source and target training images with a fusion weight.…”

Section: Related Workmentioning

confidence: 99%

FMWDCT: Foreground Mixup Into Weighted Dual-Network Cross Training for Semisupervised Remote Sensing Road Extraction

You

Wang

Chen

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

With the development of deep learning, the application of automatic road extraction has achieved great success. However, the main challenge is how to make full use of a large number of unlabeled images to improve segmentation models and how alleviate sample imbalance in road extraction tasks. In this paper, we propose a novel semi-supervised remote sensing road extraction approach is refined as Foreground Mixup into Weighted Dual-network Cross Training (FMWDCT), which combines labeled images with unlabeled images to extract road from remote sensing images. FMWDCT is composed of Dualnetwork Cross Training (DCT) and Foreground Pasting (FP). DCT is a new semi-supervised training method and FP is an effective data perturbation method for road extraction. We firstly paste the foreground pixels obtained from labeled images into unlabeled images to produce mixed input images. The mixed pseudo labels are then generated by a combination of highconfidence predictions from the augmented network and labeled masks. Finally, the mixed pseudo labels are used to guide another adversarial basic network for cross training, and this basic network is used to smoothly update the augmented network that corresponds to it. The proposed FMWDCT effectively solve the overfitting problem and imbalance problem of positive and negative sample in the case of a few labeled training samples. We demonstrate the effectiveness of our method on three road extraction datasets, and achieve better performance with few labeled data. Extensive experiments show that the proposed semi-supervised method can learn latent information from the unlabeled data to improve performance.

show abstract

Section: Related Workmentioning

confidence: 99%

FMWDCT: Foreground Mixup Into Weighted Dual-Network Cross Training for Semisupervised Remote Sensing Road Extraction

You

Wang

Chen

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

show abstract

“…To further improve the adaptation ability, some studies transfer the source images to target image previously with unsupervised manner, such as the style transfer [35]- [37]. Recent works [38], [39] try to mix up the images to provide a intermediate domain. As stated in [24], these methods mainly focus on the common knowledge and ignores the private knowledge from a certain domain.…”

Section: A Unsupervised Domain Adaptation For Semantic Segmentationmentioning

confidence: 99%

Unsupervised Domain Adaptation with Implicit Pseudo Supervision for Semantic Segmentation

Xu¹,

Wang²,

Bian³

2022

Preprint

View full text Add to dashboard Cite

Pseudo-labelling is a popular technique in unsupervised domain adaptation for semantic segmentation. However, pseudo labels are noisy and inevitably have confirmation bias due to the discrepancy between source and target domains and training process. In this paper, we train the model by the pseudo labels which are implicitly produced by itself to learn new complementary knowledge about target domain. Specifically, we propose a tri-learning architecture, where every two branches produce the pseudo labels to train the third one. And we align the pseudo labels based on the similarity of the probability distributions for each two branches. To further implicitly utilize the pseudo labels, we maximize the distances of features for different classes and minimize the distances for the same classes by triplet loss. Extensive experiments on GTA5 to Cityscapes and SYNTHIA to Cityscapes tasks show that the proposed method has considerable improvements.

show abstract

“…Since directly evaluating on one domain with a model trained on another domain usually causes performance drops, domain adaptation methods are developed to help address the domain shift between two distinct domains. Domain adaptation is widely used in classification [14,17,21,64], object detection [8,33,52,65,66], semantic segmentation [20,27,57,71,78] and many other fields [39,72,74,75]. In particular, unsupervised domain adaptation (UDA) methods have attracted substantial attention since they are free from manuallyannotated labels in the target domain.…”

Section: Domain Adaptationmentioning

confidence: 99%

Cross-Modality Domain Adaptation for Freespace Detection: A Simple yet Effective Baseline

Wang

Zhu

Huang

et al. 2022

Proceedings of the 30th ACM International Conference on Multimedia

View full text Add to dashboard Cite

As one of the fundamental functions of autonomous driving system, freespace detection aims at classifying each pixel of the image captured by the camera as drivable or non-drivable. Current works of freespace detection heavily rely on large amount of densely labeled training data for accuracy and robustness, which is time-consuming and laborious to collect and annotate. To the best of our knowledge, we are the first work to explore unsupervised domain adaptation for freespace detection to alleviate the data limitation problem with synthetic data. We develop a cross-modality domain adaptation framework which exploits both RGB images and surface normal maps generated from depth images. A Collaborative Cross Guidance (CCG) module is proposed to leverage the context information of one modality to guide the other modality in a cross manner, thus realizing inter-modality intra-domain complement. To better bridge the domain gap between source domain (synthetic data) and target domain (real-world data), we also propose a Selective Feature Alignment (SFA) module which only aligns the features of consistent foreground area between the two domains, thus realizing inter-domain intra-modality adaptation. Extensive experiments are conducted by adapting three different synthetic datasets to one real-world dataset for freespace detection respectively. Our method

show abstract

DSP: Dual Soft-Paste for Unsupervised Domain Adaptive Semantic Segmentation

Cited by 55 publications

References 42 publications

FMWDCT: Foreground Mixup Into Weighted Dual-Network Cross Training for Semisupervised Remote Sensing Road Extraction

FMWDCT: Foreground Mixup Into Weighted Dual-Network Cross Training for Semisupervised Remote Sensing Road Extraction

Unsupervised Domain Adaptation with Implicit Pseudo Supervision for Semantic Segmentation

Cross-Modality Domain Adaptation for Freespace Detection: A Simple yet Effective Baseline

Contact Info

Product

Resources

About