Continual Semantic Segmentation via Structure Preserving and Projected Feature Alignment

Lin, Zihan; Wang, Zilei; Zhang, Yixin

doi:10.1007/978-3-031-19818-2_20

Cited by 9 publications

(4 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Liu et al [52] propose a dynamic prototype convolution network by generating dynamic kernels from a support set, and then information interaction is achieved by convolution operations over query features using these kernels. Lin et al [31] disentangle the processes of retaining old knowledge and learning new classes for intermediate features, it conducts feature alignment in the encoder and calculates class prototypes in the decoder. In the remotesensing field, Li et al [102] propose a prototype update mechanism to alleviate the non-adaptive representative prototypes problem.…”

Section: Replay Methodsmentioning

confidence: 99%

“…As depicted in Fig. 3, the first category, known as data-replay methods, involves storing a portion of past training data as exemplar memory such as [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36]. The second category, termed datafree methods, includes methods like [37], [38], [39], [40], [41], [42], [43], [44], [45], [46], [47], [48], [49].…”

Section: Semantic Driftmentioning

confidence: 99%

“…ALIFE [32] proposes a featurereplay scheme that memorizes features of old classes. While SPPA [31] and SDR [27] introduce prototype alignment to CSS. Instead of directly replaying old samples, they save high-level features as the guidance of old knowledge to reduce the memory burden while utilizing the knowledge from old data.…”

Section: Qualitative Comparisonmentioning

confidence: 99%

See 2 more Smart Citations

Ecosystem Health of the Beiyun River Basin (Beijing, China) as Evaluated by the Analytic Hierarchy Process Method

Yuan

Zhao

et al. 2021

Preprint

View full text Add to dashboard Cite

Ecosystem services provided by river ecosystems rely on healthy ecosystem structure and ecological processes. The Beijing-Tianjin-Hebei urban is a typical water-deficient area. As an important part of the urban-rural integration construction, evaluating the health status of the Beiyun River Basin and discovering the weak links in the water environment is the basis for improving the health of the basin. In this study, Analytic Hierarchy Process was used to establish an evaluation index system for the Beiyun River Basin from 5 aspects including water quality, biology, habitat, hydrology, and social functions, and to assign weights to the index layer. The evaluation results showed that the health evaluation results of the Beiyun River Basin in 2019 are “sub-healthy”, and the overall health status is gradually worsening from northwest to southeast. In the middle reaches of the region, the evaluation result is "healthy", followed by the upstream, and the downstream is the worst. The results showed that areas with less human interference are in better health. The factors that affect the overall health evaluation status in the basin are the level of nutrition, biodiversity, and vegetation coverage. For the comprehensive management of the Beiyun River, the improvement of water quality and habitat ecological restoration is the key to the health of the upstream ecosystem health status. The improvement of the health status of the downstream should focus on equal emphasis on water quality and quantity, restoration of biodiversity, and improvement of the quality of the riparian ecological environment.

show abstract

Section: Replay Methodsmentioning

confidence: 99%

Section: Semantic Driftmentioning

confidence: 99%

See 1 more Smart Citation

Ecosystem Health of the Beiyun River Basin (Beijing, China) as Evaluated by the Analytic Hierarchy Process Method

Yuan

Zhao

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In recent years, RBC [53] points out that context in the CISS task is important and decouples different classes through context-rectified image-duplet learning. SPPA [54] alleviates forgetting by measuring and constraining inter-class and intra-class relationships. Incrementer [15] proposes a full Transformer framework and designs brand-new distillation and class de-confusion strategies.…”

Section: Class-incremental Semantic Segmentationmentioning

confidence: 99%

Class-Prompting Transformer for Incremental Semantic Segmentation

Song,

Shi,

Shang

et al. 2023

IEEE Access

View full text Add to dashboard Cite

Class-incremental Semantic Segmentation (CISS) aims to learn new tasks sequentially that assign a specific category to each pixel of a given image while preserving the original capability to segment the old classes even if the labels of old tasks are absent. Most existing CISS methods suppress catastrophic forgetting by directly distilling on specific layers, which ignores the semantic gap between training data from the old and new classes with different distributions and leads to distillation errors, thus affecting segmentation performance. In this paper, we propose a Class-prompting Transformer (CPT) to introduce external prior knowledge provided by a pre-trained vision-language encoder into CISS pipelines for bridging the old and new classes and performing more generalized initialization and distillation. Specifically, we proposed a Prompt-guided Initialization Module (PIM), which measures the relationships between the class prompts and old query parameters to initialize the new query parameters for relocating the previous knowledge to the learning of new tasks. Then, a Semantic-aligned Distillation Module (SDM) is proposed to incorporate class prompt information with the class-aware embeddings extracted from the decoder to prevent the semantic gap problem between distinct class data and conduct adaptive knowledge transfer to suppress catastrophic forgetting. Extensive experiments on Pascal VOC and ADE20K datasets for the CISS task demonstrate the superiority of the proposed method, which achieves state-of-the-art performance.INDEX TERMS Incremental semantic segmentation, knowledge distillation, class prompt learning.LINFENG XU (Member, IEEE) received the Ph.D. degree in signal and information processing from the School of Electronic Engineering,

show abstract

Incrementer: Transformer for Class-Incremental Semantic Segmentation with Knowledge Distillation Focusing on Old Class

Shang,

Li,

Meng

et al. 2023

2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

Continual Semantic Segmentation via Structure Preserving and Projected Feature Alignment

Cited by 9 publications

References 37 publications

Ecosystem Health of the Beiyun River Basin (Beijing, China) as Evaluated by the Analytic Hierarchy Process Method

Ecosystem Health of the Beiyun River Basin (Beijing, China) as Evaluated by the Analytic Hierarchy Process Method

Class-Prompting Transformer for Incremental Semantic Segmentation

Incrementer: Transformer for Class-Incremental Semantic Segmentation with Knowledge Distillation Focusing on Old Class

Contact Info

Product

Resources

About