Matches in SemOpenAlex for { <https://semopenalex.org/work/W3204231799> ?p ?o ?g. }
- W3204231799 abstract "Zero-Shot Learning (ZSL) targets at recognizing unseen categories byleveraging auxiliary information, such as attribute embedding. Despite theencouraging results achieved, prior ZSL approaches focus on improving thediscriminant power of seen-class features, yet have largely overlooked thegeometric structure of the samples and the prototypes. The subsequentattribute-based generative adversarial network (GAN), as a result, alsoneglects the topological information in sample generation and further yieldsinferior performances in classifying the visual features of unseen classes. Inthis paper, we introduce a novel structure-aware feature generation scheme,termed as SA-GAN, to explicitly account for the topological structure inlearning both the latent space and the generative networks. Specifically, weintroduce a constraint loss to preserve the initial geometric structure whenlearning a discriminative latent space, and carry out our GAN training withadditional supervising signals from a structure-aware discriminator and areconstruction module. The former supervision distinguishes fake and realsamples based on their affinity to class prototypes, while the latter aims toreconstruct the original feature space from the generated latent space. Thistopology-preserving mechanism enables our method to significantly enhance thegeneralization capability on unseen-classes and consequently improve theclassification performance. Experiments on four benchmarks demonstrate that theproposed approach consistently outperforms the state of the art. Our code canbe found in the supplementary material and will also be made publiclyavailable." @default.
- W3204231799 created "2021-10-11" @default.
- W3204231799 creator A5001819736 @default.
- W3204231799 creator A5015574447 @default.
- W3204231799 creator A5017818991 @default.
- W3204231799 creator A5060468648 @default.
- W3204231799 creator A5071037763 @default.
- W3204231799 date "2021-08-16" @default.
- W3204231799 modified "2023-09-24" @default.
- W3204231799 title "Structure-Aware Feature Generation for Zero-Shot Learning" @default.
- W3204231799 cites W1797268635 @default.
- W3204231799 cites W2070148066 @default.
- W3204231799 cites W2098411764 @default.
- W3204231799 cites W2099471712 @default.
- W3204231799 cites W2109317801 @default.
- W3204231799 cites W2123024445 @default.
- W3204231799 cites W2128532956 @default.
- W3204231799 cites W2134270519 @default.
- W3204231799 cites W2153579005 @default.
- W3204231799 cites W2187089797 @default.
- W3204231799 cites W2194775991 @default.
- W3204231799 cites W2289084343 @default.
- W3204231799 cites W2294130536 @default.
- W3204231799 cites W2334493732 @default.
- W3204231799 cites W2398118205 @default.
- W3204231799 cites W2467604901 @default.
- W3204231799 cites W2520774990 @default.
- W3204231799 cites W2533598788 @default.
- W3204231799 cites W2552383788 @default.
- W3204231799 cites W2554246103 @default.
- W3204231799 cites W2571899125 @default.
- W3204231799 cites W2596142952 @default.
- W3204231799 cites W2611632661 @default.
- W3204231799 cites W27961112 @default.
- W3204231799 cites W2883334176 @default.
- W3204231799 cites W2891932073 @default.
- W3204231799 cites W2917742317 @default.
- W3204231799 cites W2924476266 @default.
- W3204231799 cites W2950276680 @default.
- W3204231799 cites W2950577311 @default.
- W3204231799 cites W2951004968 @default.
- W3204231799 cites W2955087950 @default.
- W3204231799 cites W2962716452 @default.
- W3204231799 cites W2962824231 @default.
- W3204231799 cites W2962830213 @default.
- W3204231799 cites W2963283377 @default.
- W3204231799 cites W2963499153 @default.
- W3204231799 cites W2963960318 @default.
- W3204231799 cites W2964086552 @default.
- W3204231799 cites W2964307109 @default.
- W3204231799 cites W2970733445 @default.
- W3204231799 cites W2979571231 @default.
- W3204231799 cites W2979728922 @default.
- W3204231799 cites W2981407849 @default.
- W3204231799 cites W2986692991 @default.
- W3204231799 cites W2990947836 @default.
- W3204231799 cites W2998708909 @default.
- W3204231799 cites W3000538487 @default.
- W3204231799 cites W3015970443 @default.
- W3204231799 cites W3034359780 @default.
- W3204231799 cites W3034379915 @default.
- W3204231799 cites W3035297600 @default.
- W3204231799 cites W3099554308 @default.
- W3204231799 cites W3109728025 @default.
- W3204231799 cites W652269744 @default.
- W3204231799 cites W93016980 @default.
- W3204231799 hasPublicationYear "2021" @default.
- W3204231799 type Work @default.
- W3204231799 sameAs 3204231799 @default.
- W3204231799 citedByCount "0" @default.
- W3204231799 crossrefType "posted-content" @default.
- W3204231799 hasAuthorship W3204231799A5001819736 @default.
- W3204231799 hasAuthorship W3204231799A5015574447 @default.
- W3204231799 hasAuthorship W3204231799A5017818991 @default.
- W3204231799 hasAuthorship W3204231799A5060468648 @default.
- W3204231799 hasAuthorship W3204231799A5071037763 @default.
- W3204231799 hasConcept C111919701 @default.
- W3204231799 hasConcept C114614502 @default.
- W3204231799 hasConcept C119857082 @default.
- W3204231799 hasConcept C120665830 @default.
- W3204231799 hasConcept C121332964 @default.
- W3204231799 hasConcept C134306372 @default.
- W3204231799 hasConcept C138885662 @default.
- W3204231799 hasConcept C153180895 @default.
- W3204231799 hasConcept C154945302 @default.
- W3204231799 hasConcept C177148314 @default.
- W3204231799 hasConcept C184720557 @default.
- W3204231799 hasConcept C192209626 @default.
- W3204231799 hasConcept C2524010 @default.
- W3204231799 hasConcept C2776036281 @default.
- W3204231799 hasConcept C2776401178 @default.
- W3204231799 hasConcept C2777212361 @default.
- W3204231799 hasConcept C2778572836 @default.
- W3204231799 hasConcept C2779803651 @default.
- W3204231799 hasConcept C33923547 @default.
- W3204231799 hasConcept C39890363 @default.
- W3204231799 hasConcept C41008148 @default.
- W3204231799 hasConcept C41608201 @default.
- W3204231799 hasConcept C41895202 @default.
- W3204231799 hasConcept C59404180 @default.