Shaping Visual Representations With Attributes for Few-Shot Recognition

Chen, Haoxing; Li, Huaxiong; Li, Yaohui; Chen, Chunlin

doi:10.1109/lsp.2022.3180934

Cited by 6 publications

(2 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our experiments are conducted using the convolution neural network ResNet12 (Chen et al 2021b), a popular feature extractor in recent few-shot learning methods. We also give our results using Conv4 (Vinyals et al 2016) on SUN for fair comparisons since we find that almost all previous works (Huang et al 2021;Ji et al 2022;Chen et al 2022;Xing et al 2019) choose Conv4 as their feature extractor on SUN. Moreover, we use a simple fullyconnected layer as the learnable weight generator G, and the temperature parameter τ 1 and τ 2 are initialized as 10.…”

Section: Experiments Experimental Setupmentioning

confidence: 99%

“…Based on the above analysis, our motivation is to explicitly learn some fine-grained and transferable metaknowledge from base classes and then reuse the learned meta-knowledge to recognize novel classes with only a few labeled examples. As some previous works (Huang et al 2021;Tokmakov, Wang, and Hebert 2019;Ji et al 2022;Xing et al 2019;Chen et al 2022), our work also utilizes category-level attribute annotations, i.e., only one attribute score vector for each class. AM3 (Xing et al 2019) proposes a modality mixture mechanism that can adaptively combine visual and semantic information.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Compositional Prototypical Networks for Few-Shot Classification

Lyu

Wang

2023

AAAI

View full text Add to dashboard Cite

It is assumed that pre-training provides the feature extractor with strong class transferability and that high novel class generalization can be achieved by simply reusing the transferable feature extractor. In this work, our motivation is to explicitly learn some fine-grained and transferable meta-knowledge so that feature reusability can be further improved. Concretely, inspired by the fact that humans can use learned concepts or components to help them recognize novel classes, we propose Compositional Prototypical Networks (CPN) to learn a transferable prototype for each human-annotated attribute, which we call a component prototype. We empirically demonstrate that the learned component prototypes have good class transferability and can be reused to construct compositional prototypes for novel classes. Then a learnable weight generator is utilized to adaptively fuse the compositional and visual prototypes. Extensive experiments demonstrate that our method can achieve state-of-the-art results on different datasets and settings. The performance gains are especially remarkable in the 5-way 1-shot setting. The code is available at https://github.com/fikry102/CPN.

show abstract