Multi-Prototype Few-shot Learning in Histopathology

Deuschel, Jessica; Firmbach, Daniel; Geppert, Carol; Eckstein, Markus; Hartmann, Arndt; Bruns, Volker; Kuritcyn, Petr; Dexl, Jakob; Hartmann, David; Perrin, Dominik; Wittenberg, Thomas; Benz, Michaela

doi:10.1109/iccvw54120.2021.00075

Cited by 11 publications

(2 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Originated in few-shot learning, multi-prototype learning aims to address the challenging of fitting prototypical network [25] for multi-modal data distribution [5], [26], [27] by learning prototypes for recognizing classes with few training examples. The first attempt, IMP [5], proposed to adaptively expand prototype pool following a Chinese restaurant process which sequentially processes data points.…”

Section: Multi-prototype Learningmentioning

confidence: 99%

“…The first attempt, IMP [5], proposed to adaptively expand prototype pool following a Chinese restaurant process which sequentially processes data points. [27] proposes a kmeans extension of Prototypical Networks. Despite the success in few-shot learning, it is impractical to trivially apply the sequential IMP to weakly supervised learning due to computation cost while other offline methods prevents endto-end training of prototypes.…”

Section: Multi-prototype Learningmentioning

confidence: 99%

See 1 more Smart Citation

Weakly Supervised 3D Point Cloud Segmentation via Multi-Prototype Learning

Su¹,

Xu²,

Jia³

2022

Preprint

View full text Add to dashboard Cite

Addressing the annotation challenge in 3D Point Cloud segmentation has inspired research into weakly supervised learning. Existing approaches mainly focus on exploiting manifold and pseudo-labeling to make use of large unlabeled data points. A fundamental challenge here lies in the large intra-class variations of local geometric structure, resulting in subclasses within a semantic class. In this work, we leverage this intuition and opt for maintaining an individual classifier for each subclass. Technically, we design a multi-prototype classifier, each prototype serves as the classifier weights for one subclass. To enable effective updating of multi-prototype classifier weights, we propose two constraints respectively for updating the prototypes w.r.t. all point features and for encouraging the learning of diverse prototypes. Experiments on weakly supervised 3D point cloud segmentation tasks validate the efficacy of proposed method in particular at low-label regime. Our hypothesis is also verified given the consistent discovery of semantic subclasses at no cost of additional annotations.

show abstract

Section: Multi-prototype Learningmentioning

confidence: 99%

Section: Multi-prototype Learningmentioning

confidence: 99%