Unsupervised Intra-Domain Adaptation for Semantic Segmentation Through Self-Supervision

Pan, Fang; Shin, Inkyu; Rameau, François; Lee, Seokju; Kweon, In So

doi:10.1109/cvpr42600.2020.00382

Cited by 347 publications

(285 citation statements)

References 28 publications

Supporting

Mentioning

271

Contrasting

Order By: Relevance

“…The easy-to-adapt images are then used to aid the adaptation of the hard-toadapt images. Although this second-stage adaptation (which is called intra-domain adaptation in [18]) indeed improves the overall predication, we found that this image-level splitting fails to yield accurate prediction for certain target pixels. Especially, because not all the target pixels in the easy-to-adapt split are equally trustworthy, this image-level split may not be a good solution to deal with the variant target data.…”

Section: Introductionmentioning

confidence: 82%

“…1 (d). Similar concerns have been raised in [18], where all the target images are classified into two categories, i.e., an easyto-adapt split and a hard-to-adapt split, according to the overall prediction confidence of each image. The easy-to-adapt images are then used to aid the adaptation of the hard-toadapt images.…”

Section: Introductionmentioning

confidence: 85%

“…Furthermore, the notable class-imbalanced issue in image semantic segmentation also needs to be taken good care of. As mentioned in [18], those images containing rare classes (e.g., "fence", "light", and "sign") are usually classified into the hard-to-adapt split. Because the easy-to-adapt images themselves have few rare-class samples (even if they do, the rare-class samples tend to be wrongly predicted) and are not reliable to guide the others, the class-imbalanced situation may further deteriorate after the adaptation.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Self-Guided Adversarial Learning For Domain Adaptive Semantic Segmentation

Pang¹,

Chang²,

Hsu

2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

Unsupervised domain adaptation has been introduced to generalize semantic segmentation models from labeled synthetic images to unlabeled real-world images. Although much effort was devoted to minimize the cross-domain gap, the segmentation results on real-world data remain highly unstable. In this paper, we discuss two main issues which hinder previous methods from achieving satisfactory results and propose a novel self-guided adversarial learning to leverage the capability of domain adaptation. Firstly, to deal with the unpredictable data variation in the real-world domain, we develop a self-guided adversarial learning method by selecting reliable target pixels as guidance to lead the adaptation of the other pixels. Secondly, to address the class-imbalanced issue, we devise the selection strategy in each class independently and incorporate this idea with a class-level adversarial learning in a unified framework. Experimental results show that the proposed method significantly improves the previous methods on several benchmark datasets.

show abstract

Section: Introductionmentioning

confidence: 82%

Section: Introductionmentioning

confidence: 85%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Self-Guided Adversarial Learning For Domain Adaptive Semantic Segmentation

Pang¹,

Chang²,

Hsu

2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

show abstract

“…Indeed, this method has a positive performance on the nighttime image segmentation, but the establishment of a fine annotation dataset costs much manpower and material resources. Translating the target domain data to the source domain and the production of the pseudo labels or weak labels for unlabeld target data are proposed for alleviating the burden of specific situation dataset creation [8,9,10,11]. There are part of solutions for model optimization that concentrate on the narrowing the gap of the feature between source and target as well [12,13].…”

Section: Our Methodsmentioning

confidence: 99%

Semantic Nighttime Image Segmentation Via Illumination and Position Aware Domain Adaptation

Peng¹,

Su²,

Sun³

et al. 2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

Due to the lack of the annotated nighttime images, general image segmentation models trained on the daytime image dataset do not perform well in nighttime scenes. The difference of the illumination condition and the difficulty to obtain the position information between daytime and nighttime makes the nighttime image segmentation tough. As a consequence, this paper proposes an end-to-end nighttime segmentation network based on the following two points: 1) Utilizing illumination adaptation with the different illumination condition on the daytime or nighttime to close the distribution gap at the feature map level; 2) With the prior information about the position of each object in the outdoor scene, some classification errors could be corrected by incorporating the self-attention mechanism. The scheme is tested on the opensource nighttime dataset Dark Zurich and night driving, with a 2.5% improvement compared to the base segmentation network.

show abstract

“…Semantic segmentation aims at predicting a class label for each pixel in the image, which plays a crucial role in various applications, including autonomous driving [1][2][3], robotics [4,5], medical applications [6], and augmented reality [7]. Because of the success of CNN in recent years, a large number of semantic segmentation algorithms based on deep learning have been proposed, which have made a breakthrough in this filed [8][9][10][11].…”

Section: Introductionmentioning

confidence: 99%

Semantic Segmentation Leveraging Simultaneous Depth Estimation

Sun

Gao

Cui

et al. 2021

Sensors

View full text Add to dashboard Cite

Semantic segmentation is one of the most widely studied problems in computer vision communities, which makes a great contribution to a variety of applications. A lot of learning-based approaches, such as Convolutional Neural Network (CNN), have made a vast contribution to this problem. While rich context information of the input images can be learned from multi-scale receptive fields by convolutions with deep layers, traditional CNNs have great difficulty in learning the geometrical relationship and distribution of objects in the RGB image due to the lack of depth information, which may lead to an inferior segmentation quality. To solve this problem, we propose a method that improves segmentation quality with depth estimation on RGB images. Specifically, we estimate depth information on RGB images via a depth estimation network, and then feed the depth map into the CNN which is able to guide the semantic segmentation. Furthermore, in order to parse the depth map and RGB images simultaneously, we construct a multi-branch encoder–decoder network and fuse the RGB and depth features step by step. Extensive experimental evaluation on four baseline networks demonstrates that our proposed method can enhance the segmentation quality considerably and obtain better performance compared to other segmentation networks.

show abstract

Unsupervised Intra-Domain Adaptation for Semantic Segmentation Through Self-Supervision

Cited by 347 publications

References 28 publications

Self-Guided Adversarial Learning For Domain Adaptive Semantic Segmentation

Self-Guided Adversarial Learning For Domain Adaptive Semantic Segmentation

Semantic Nighttime Image Segmentation Via Illumination and Position Aware Domain Adaptation

Semantic Segmentation Leveraging Simultaneous Depth Estimation

Contact Info

Product

Resources

About