Matches in SemOpenAlex for { <https://semopenalex.org/work/W3210458440> ?p ?o ?g. }
Showing items 1 to 85 of
85
with 100 items per page.
- W3210458440 abstract "Vision-language transformers (VL transformers) have shown impressive accuracyin cross-modal retrieval. However, most of the existing VL transformers useearly-interaction dataflow that computes a joint representation for thetext-image input. In the retrieval stage, such models need to infer on all thematched text-image combinations, which causes high computing costs. The goal ofthis paper is to decompose the early-interaction dataflow inside thepre-trained VL transformer to achieve acceleration while maintaining itsoutstanding accuracy. To achieve this, we propose a novel Vision-languageTransformer Decomposing (VLDeformer) to modify the VL transformer as anindividual encoder for a single image or text through contrastive learning,which accelerates retrieval speed by thousands of times. Meanwhile, we proposeto compose bi-modal hard negatives for the contrastive learning objective,which enables the VLDeformer to maintain the outstanding accuracy of thebackbone VL transformer. Extensive experiments on COCO and Flickr30k datasetsdemonstrate the superior performance of the proposed method. Considering botheffectiveness and efficiency, VLDeformer provides a superior selection forcross-modal retrieval in the similar pre-training datascale." @default.
- W3210458440 created "2021-11-08" @default.
- W3210458440 creator A5000839824 @default.
- W3210458440 creator A5008740845 @default.
- W3210458440 creator A5041123189 @default.
- W3210458440 creator A5049854957 @default.
- W3210458440 creator A5057622363 @default.
- W3210458440 creator A5060298257 @default.
- W3210458440 creator A5060382408 @default.
- W3210458440 creator A5070202314 @default.
- W3210458440 creator A5072624830 @default.
- W3210458440 date "2021-10-20" @default.
- W3210458440 modified "2023-09-28" @default.
- W3210458440 title "VLDeformer: Learning Visual-Semantic Embeddings by Vision-Language Transformer Decomposing" @default.
- W3210458440 hasPublicationYear "2021" @default.
- W3210458440 type Work @default.
- W3210458440 sameAs 3210458440 @default.
- W3210458440 citedByCount "0" @default.
- W3210458440 crossrefType "posted-content" @default.
- W3210458440 hasAuthorship W3210458440A5000839824 @default.
- W3210458440 hasAuthorship W3210458440A5008740845 @default.
- W3210458440 hasAuthorship W3210458440A5041123189 @default.
- W3210458440 hasAuthorship W3210458440A5049854957 @default.
- W3210458440 hasAuthorship W3210458440A5057622363 @default.
- W3210458440 hasAuthorship W3210458440A5060298257 @default.
- W3210458440 hasAuthorship W3210458440A5060382408 @default.
- W3210458440 hasAuthorship W3210458440A5070202314 @default.
- W3210458440 hasAuthorship W3210458440A5072624830 @default.
- W3210458440 hasConcept C111919701 @default.
- W3210458440 hasConcept C118505674 @default.
- W3210458440 hasConcept C119599485 @default.
- W3210458440 hasConcept C127413603 @default.
- W3210458440 hasConcept C154945302 @default.
- W3210458440 hasConcept C165801399 @default.
- W3210458440 hasConcept C185592680 @default.
- W3210458440 hasConcept C188027245 @default.
- W3210458440 hasConcept C199360897 @default.
- W3210458440 hasConcept C204321447 @default.
- W3210458440 hasConcept C28490314 @default.
- W3210458440 hasConcept C41008148 @default.
- W3210458440 hasConcept C66322947 @default.
- W3210458440 hasConcept C71139939 @default.
- W3210458440 hasConcept C96324660 @default.
- W3210458440 hasConceptScore W3210458440C111919701 @default.
- W3210458440 hasConceptScore W3210458440C118505674 @default.
- W3210458440 hasConceptScore W3210458440C119599485 @default.
- W3210458440 hasConceptScore W3210458440C127413603 @default.
- W3210458440 hasConceptScore W3210458440C154945302 @default.
- W3210458440 hasConceptScore W3210458440C165801399 @default.
- W3210458440 hasConceptScore W3210458440C185592680 @default.
- W3210458440 hasConceptScore W3210458440C188027245 @default.
- W3210458440 hasConceptScore W3210458440C199360897 @default.
- W3210458440 hasConceptScore W3210458440C204321447 @default.
- W3210458440 hasConceptScore W3210458440C28490314 @default.
- W3210458440 hasConceptScore W3210458440C41008148 @default.
- W3210458440 hasConceptScore W3210458440C66322947 @default.
- W3210458440 hasConceptScore W3210458440C71139939 @default.
- W3210458440 hasConceptScore W3210458440C96324660 @default.
- W3210458440 hasLocation W32104584401 @default.
- W3210458440 hasOpenAccess W3210458440 @default.
- W3210458440 hasPrimaryLocation W32104584401 @default.
- W3210458440 hasRelatedWork W2760575525 @default.
- W3210458440 hasRelatedWork W3016663370 @default.
- W3210458440 hasRelatedWork W3029678209 @default.
- W3210458440 hasRelatedWork W3110662498 @default.
- W3210458440 hasRelatedWork W3128099838 @default.
- W3210458440 hasRelatedWork W3128723389 @default.
- W3210458440 hasRelatedWork W3151130473 @default.
- W3210458440 hasRelatedWork W3165647589 @default.
- W3210458440 hasRelatedWork W3166658420 @default.
- W3210458440 hasRelatedWork W3167695527 @default.
- W3210458440 hasRelatedWork W3168491317 @default.
- W3210458440 hasRelatedWork W3175466730 @default.
- W3210458440 hasRelatedWork W3181262653 @default.
- W3210458440 hasRelatedWork W3199613405 @default.
- W3210458440 hasRelatedWork W3208383346 @default.
- W3210458440 hasRelatedWork W3209108016 @default.
- W3210458440 hasRelatedWork W3210358254 @default.
- W3210458440 hasRelatedWork W3211483028 @default.
- W3210458440 hasRelatedWork W3212832343 @default.
- W3210458440 hasRelatedWork W3121735241 @default.
- W3210458440 isParatext "false" @default.
- W3210458440 isRetracted "false" @default.
- W3210458440 magId "3210458440" @default.
- W3210458440 workType "article" @default.