论文阅读 [TPAMI-2022] Prior Guided Feature Enrichment Network for Few-Shot Segmentation
论文搜索(studyai.com)
搜索论文: Prior Guided Feature Enrichment Network for Few-Shot Segmentation
搜索论文: http://www.studyai.com/search/whole-site/?q=Prior+Guided+Feature+Enrichment+Network+for+Few-Shot+Segmentation
关键字(Keywords)
Semantics; Image segmentation; Object segmentation; Training; Finite element analysis; Adaptation models; Feature extraction; Few-shot segmentation; few-shot learning; semantic segmentation; scene understanding
机器学习; 机器视觉
检测分割; 小样本学习; 场景解析
摘要(Abstract)
State-of-the-art semantic segmentation methods require sufficient labeled data to achieve good results and hardly work on unseen classes without fine-tuning.
最先进的语义分割方法需要足够的标记数据才能获得良好的结果,而且如果不进行微调,就很难处理看不见的类。.
Few-shot segmentation is thus proposed to tackle this problem by learning a model that quickly adapts to new classes with a few labeled support samples.
因此,提出了一种通过学习一个模型来解决这个问题的少量镜头分割方法,该模型可以快速适应具有少量标记支持样本的新类。.
Theses frameworks still face the challenge of generalization ability reduction on unseen classes due to inappropriate use of high-level semantic information of training classes and spatial inconsistency between query and support targets.
由于不恰当地使用训练类的高级语义信息,以及查询和支持目标之间的空间不一致性,这些框架仍然面临着对看不见类的泛化能力降低的挑战。.
To alleviate these issues, we propose the Prior Guided Feature Enrichment Network (PFENet).
为了缓解这些问题,我们提出了先验引导特征丰富网络(PFENet)。.
It consists of novel designs of (1) a training-free prior mask generation method that not only retains generalization power but also improves model performance and (2) Feature Enrichment Module (FEM) that overcomes spatial inconsistency by adaptively enriching query features with support features and prior masks.
它包括:(1)一种无训练的先验掩码生成方法,该方法不仅保留了泛化能力,而且提高了模型性能;(2)特征丰富模块(FEM),该模块通过支持特征和先验掩码自适应地丰富查询特征,克服了空间不一致性。.
Extensive experiments on PASCAL-5 i ^i ii and COCO prove that the proposed prior generation method and FEM both improve the baseline method significantly.
在PASCAL-5 i ^i ii和COCO上进行的大量实验证明,所提出的前代方法和FEM都显著改善了基线方法。.
Our PFENet also outperforms state-of-the-art methods by a large margin without efficiency loss.
我们的PFENet在没有效率损失的情况下,大大优于最先进的方法。.
It is surprising that our model even generalizes to cases without labeled support samples…
令人惊讶的是,我们的模型甚至可以推广到没有标记支持样本的情况。。.
作者(Authors)
[‘Zhuotao Tian’, ‘Hengshuang Zhao’, ‘Michelle Shu’, ‘Zhicheng Yang’, ‘Ruiyu Li’, ‘Jiaya Jia’]