Contextual-Relation Consistent Domain Adaptation for Semantic Segmentation

Huang, Jiaxing; Lu, Shijian; Guan, Dayan; Zhang, Xiaobing

doi:10.1007/978-3-030-58555-6_42

Cited by 100 publications

(73 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(13) Baseline. The previous cross-domain semantic segmentation methods are mainly divided into three types: 1) methods based on adversarial training [21][22][23][24][25]; 2) methods based on self-training [3] [26] [27]; and 3) methods based on data augmentation [28,29]. In order to verify the effectiveness of our proposed model, we select the representative schemes of the above three methods in recent years for comparison.…”

Section: Methodsmentioning

confidence: 99%

Boundary Enhancement and Contrastive Alignment for Unsupervised Domain Adaptive Semantic Segmentation

Líu

et al. 2022

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

In the actual construction process, the supervision work of concrete pouring has many problems, such as heavy workload, low efficiency, misjudgment and omission. Deep learning shows good performance in computer vision, such as semantic segmentation and object recognition. In this paper, semantic segmentation is used to identify the position of vibrating bar in concrete pouring to provide a basis for detecting whether the vibrating behavior is standardized. Existing semantic segmentation studies ignore whether the edges of objects are finely detected. Recently, contrastive learning has made progress in computer vision. In addition, the training data differs greatly from the actual construction scene, namely domain shift. Therefore, we proposed the domain-adaptation method BECA which consists of two parts: boundary enhancement for accurate detection of edges and contrastive alignment for domain shift. Experiments show that the proposed BECA has unique advantages compared with the previous methods.

show abstract

Section: Methodsmentioning

confidence: 99%

Boundary Enhancement and Contrastive Alignment for Unsupervised Domain Adaptive Semantic Segmentation

Líu

et al. 2022

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

show abstract

“…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]. In this paper, our team proposes an improved semantic segmentation network to optimize the performance of the images segmentation in the nighttime situation.…”

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

“…He et al [46] apply KD on a compact student network to address the mentioned problem and take advantage of a pretrained autoencoder for feature similarity optimization. There also exist other variations of single teacher KD algorithms, which propose novel approaches to distilling knowledge, such as intra-class feature variation [49] or contextualrelation consistent domain adaptation [50].…”

Section: A Knowledge Distillationmentioning

confidence: 99%

Robust Semantic Segmentation With Multi-Teacher Knowledge Distillation

Amirkhani¹,

Khosravian

Masih‐Tehrani

et al. 2021

IEEE Access

View full text Add to dashboard Cite

Recent studies have recently exploited knowledge distillation (KD) technique to address timeconsuming annotation task in semantic segmentation, through which one teacher trained on a single dataset could be leveraged for annotating unlabeled data. However, in this context, knowledge capacity is restricted, and knowledge variety is rare in different conditions, such as cross-model KD, in which the single teacher KD prohibits the student model from distilling information using cross-domain context. To fix this concern, we have looked into learning a lightweight student from a group of teachers. To be more specific, we train five distinct lightweight convolutional neural networks (CNNs) for semantic segmentation on several datasets. Several state-of-the-art augmentation transformations have also been utilized in our training phase. The impacts of such training scenarios are then assessed in terms of student robustness and accuracy. As the main contribution of this paper, our proposed multi-teacher KD paradigm endows the student with the ability to amalgamate and capture a variety of knowledge illustrations from different sources. Results demonstrated that our method outperforms the existing studies on both clean and corrupted data in the semantic segmentation task while benefiting from our proposed score weight system. Experiments validate that our multi-teacher framework results in an improvement of 9% up to 32.18% compared to the singleteacher paradigm. Moreover, it is demonstrated that our paradigm surpasses previous supervised real-time studies in the semantic segmentation challenge.

show abstract

Contextual-Relation Consistent Domain Adaptation for Semantic Segmentation

Cited by 100 publications

References 48 publications

Boundary Enhancement and Contrastive Alignment for Unsupervised Domain Adaptive Semantic Segmentation

Boundary Enhancement and Contrastive Alignment for Unsupervised Domain Adaptive Semantic Segmentation

Semantic Nighttime Image Segmentation Via Illumination and Position Aware Domain Adaptation

Robust Semantic Segmentation With Multi-Teacher Knowledge Distillation

Contact Info

Product

Resources

About