Matches in SemOpenAlex for { <https://semopenalex.org/work/W4386066387> ?p ?o ?g. }
Showing items 1 to 97 of
97
with 100 items per page.
- W4386066387 abstract "Human spatial attention conveys information about the regions of visual scenes that are important for performing visual tasks. Prior work has shown that the information about human attention can be leveraged to benefit various supervised vision tasks. Might providing this weak form of supervision be useful for self-supervised representation learning? Addressing this question requires collecting large datasets with human attention labels. Yet, collecting such large scale data is very expensive. To address this challenge, we construct an auxiliary teacher model to predict human attention, trained on a relatively small labeled dataset. This teacher model allows us to generate image (pseudo) attention labels for ImageNet. We then train a model with a primary contrastive objective; to this standard configuration, we add a simple output head trained to predict the attention map for each image, guided by the pseudo labels from teacher model. We measure the quality of learned representations by evaluating classification performance from the frozen learned embeddings as well as performance on image retrieval tasks (see supplementary material). We find that the spatial-attention maps predicted from the contrastive model trained with teacher guidance aligns better with human attention compared to vanilla contrastive models. Moreover, we find that our approach improves classification accuracy and robustness of the contrastive models on ImageNet and ImageNet-C. Further, we find that model representations become more useful for image retrieval task as measured by precision-recall performance on ImageNet, ImageNet-C, CIFAR10, and CIFAR10-C datasets." @default.
- W4386066387 created "2023-08-23" @default.
- W4386066387 creator A5010854091 @default.
- W4386066387 creator A5043002930 @default.
- W4386066387 creator A5076084047 @default.
- W4386066387 creator A5086195538 @default.
- W4386066387 date "2023-06-01" @default.
- W4386066387 modified "2023-09-27" @default.
- W4386066387 title "Teacher-generated spatial-attention labels boost robustness and accuracy of contrastive models" @default.
- W4386066387 cites W1497599070 @default.
- W4386066387 cites W1934890906 @default.
- W4386066387 cites W2030031014 @default.
- W4386066387 cites W2068419217 @default.
- W4386066387 cites W2117539524 @default.
- W4386066387 cites W2128272608 @default.
- W4386066387 cites W2144764737 @default.
- W4386066387 cites W2151167738 @default.
- W4386066387 cites W2212216676 @default.
- W4386066387 cites W2295107390 @default.
- W4386066387 cites W2470347484 @default.
- W4386066387 cites W2738450183 @default.
- W4386066387 cites W2962884579 @default.
- W4386066387 cites W2963495494 @default.
- W4386066387 cites W2963503775 @default.
- W4386066387 cites W2977639983 @default.
- W4386066387 cites W2981929423 @default.
- W4386066387 cites W2983256121 @default.
- W4386066387 cites W3005401190 @default.
- W4386066387 cites W3035160371 @default.
- W4386066387 cites W3035524453 @default.
- W4386066387 cites W3040102868 @default.
- W4386066387 cites W3086471466 @default.
- W4386066387 cites W3099512298 @default.
- W4386066387 cites W3101840568 @default.
- W4386066387 cites W3132652652 @default.
- W4386066387 cites W3168796319 @default.
- W4386066387 cites W3171007011 @default.
- W4386066387 cites W3199841966 @default.
- W4386066387 cites W3202258049 @default.
- W4386066387 cites W4224281996 @default.
- W4386066387 doi "https://doi.org/10.1109/cvpr52729.2023.02230" @default.
- W4386066387 hasPublicationYear "2023" @default.
- W4386066387 type Work @default.
- W4386066387 citedByCount "0" @default.
- W4386066387 crossrefType "proceedings-article" @default.
- W4386066387 hasAuthorship W4386066387A5010854091 @default.
- W4386066387 hasAuthorship W4386066387A5043002930 @default.
- W4386066387 hasAuthorship W4386066387A5076084047 @default.
- W4386066387 hasAuthorship W4386066387A5086195538 @default.
- W4386066387 hasConcept C100660578 @default.
- W4386066387 hasConcept C104317684 @default.
- W4386066387 hasConcept C115961682 @default.
- W4386066387 hasConcept C119857082 @default.
- W4386066387 hasConcept C138885662 @default.
- W4386066387 hasConcept C153180895 @default.
- W4386066387 hasConcept C154945302 @default.
- W4386066387 hasConcept C185592680 @default.
- W4386066387 hasConcept C199360897 @default.
- W4386066387 hasConcept C2780801425 @default.
- W4386066387 hasConcept C41008148 @default.
- W4386066387 hasConcept C41895202 @default.
- W4386066387 hasConcept C55493867 @default.
- W4386066387 hasConcept C63479239 @default.
- W4386066387 hasConcept C75294576 @default.
- W4386066387 hasConcept C81669768 @default.
- W4386066387 hasConceptScore W4386066387C100660578 @default.
- W4386066387 hasConceptScore W4386066387C104317684 @default.
- W4386066387 hasConceptScore W4386066387C115961682 @default.
- W4386066387 hasConceptScore W4386066387C119857082 @default.
- W4386066387 hasConceptScore W4386066387C138885662 @default.
- W4386066387 hasConceptScore W4386066387C153180895 @default.
- W4386066387 hasConceptScore W4386066387C154945302 @default.
- W4386066387 hasConceptScore W4386066387C185592680 @default.
- W4386066387 hasConceptScore W4386066387C199360897 @default.
- W4386066387 hasConceptScore W4386066387C2780801425 @default.
- W4386066387 hasConceptScore W4386066387C41008148 @default.
- W4386066387 hasConceptScore W4386066387C41895202 @default.
- W4386066387 hasConceptScore W4386066387C55493867 @default.
- W4386066387 hasConceptScore W4386066387C63479239 @default.
- W4386066387 hasConceptScore W4386066387C75294576 @default.
- W4386066387 hasConceptScore W4386066387C81669768 @default.
- W4386066387 hasLocation W43860663871 @default.
- W4386066387 hasOpenAccess W4386066387 @default.
- W4386066387 hasPrimaryLocation W43860663871 @default.
- W4386066387 hasRelatedWork W133358225 @default.
- W4386066387 hasRelatedWork W1577137544 @default.
- W4386066387 hasRelatedWork W2041525275 @default.
- W4386066387 hasRelatedWork W2056016498 @default.
- W4386066387 hasRelatedWork W2508908072 @default.
- W4386066387 hasRelatedWork W2509146328 @default.
- W4386066387 hasRelatedWork W2537156416 @default.
- W4386066387 hasRelatedWork W2742991909 @default.
- W4386066387 hasRelatedWork W2996038082 @default.
- W4386066387 hasRelatedWork W7626849 @default.
- W4386066387 isParatext "false" @default.
- W4386066387 isRetracted "false" @default.
- W4386066387 workType "article" @default.