FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3211882289> ?p ?o ?g. }

• W3211882289 abstract "The goal of self-supervised visual representation learning is to learnstrong, transferable image representations, with the majority of researchfocusing on object or scene level. On the other hand, representation learningat part level has received significantly less attention. In this paper, wepropose an unsupervised approach to object part discovery and segmentation andmake three contributions. First, we construct a proxy task through a set ofobjectives that encourages the model to learn a meaningful decomposition of theimage into its parts. Secondly, prior work argues for reconstructing orclustering pre-computed features as a proxy to parts; we show empirically thatthis alone is unlikely to find meaningful parts; mainly because of their lowresolution and the tendency of classification networks to spatially smear outinformation. We suggest that image reconstruction at the level of pixels canalleviate this problem, acting as a complementary cue. Lastly, we show that thestandard evaluation based on keypoint regression does not correlate well withsegmentation quality and thus introduce different metrics, NMI and ARI, thatbetter characterize the decomposition of objects into parts. Our method yieldssemantic parts which are consistent across fine-grained but visually distinctcategories, outperforming the state of the art on three benchmark datasets.Code is available at the project page:https://www.robots.ox.ac.uk/~vgg/research/unsup-parts/." @default.
• W3211882289 created "2021-11-22" @default.
• W3211882289 creator A5000810083 @default.
• W3211882289 creator A5060511349 @default.
• W3211882289 creator A5062808708 @default.
• W3211882289 creator A5083153177 @default.
• W3211882289 date "2021-11-11" @default.
• W3211882289 modified "2023-09-27" @default.
• W3211882289 title "Unsupervised Part Discovery from Contrastive Reconstruction" @default.
• W3211882289 cites W1496650988 @default.
• W3211882289 cites W1797268635 @default.
• W3211882289 cites W1898560071 @default.
• W3211882289 cites W1903370114 @default.
• W3211882289 cites W1928906481 @default.
• W3211882289 cites W1959608418 @default.
• W3211882289 cites W2036989445 @default.
• W3211882289 cites W2040553845 @default.
• W3211882289 cites W2079789819 @default.
• W3211882289 cites W2104408738 @default.
• W3211882289 cites W2104657103 @default.
• W3211882289 cites W2117539524 @default.
• W3211882289 cites W2123081785 @default.
• W3211882289 cites W2135706578 @default.
• W3211882289 cites W2138621090 @default.
• W3211882289 cites W2152826865 @default.
• W3211882289 cites W2168356304 @default.
• W3211882289 cites W2194775991 @default.
• W3211882289 cites W2226771013 @default.
• W3211882289 cites W2309415944 @default.
• W3211882289 cites W2321533354 @default.
• W3211882289 cites W2326925005 @default.
• W3211882289 cites W2412782625 @default.
• W3211882289 cites W2471768434 @default.
• W3211882289 cites W2479109623 @default.
• W3211882289 cites W2527787998 @default.
• W3211882289 cites W2552264258 @default.
• W3211882289 cites W2741910023 @default.
• W3211882289 cites W2773003563 @default.
• W3211882289 cites W2798365843 @default.
• W3211882289 cites W2807242871 @default.
• W3211882289 cites W2807931652 @default.
• W3211882289 cites W2842511635 @default.
• W3211882289 cites W2883725317 @default.
• W3211882289 cites W2887997457 @default.
• W3211882289 cites W2891951760 @default.
• W3211882289 cites W2903585023 @default.
• W3211882289 cites W2907236913 @default.
• W3211882289 cites W2911448865 @default.
• W3211882289 cites W2944828972 @default.
• W3211882289 cites W2952484912 @default.
• W3211882289 cites W2962742544 @default.
• W3211882289 cites W2962784289 @default.
• W3211882289 cites W2962835968 @default.
• W3211882289 cites W2962981304 @default.
• W3211882289 cites W2963203586 @default.
• W3211882289 cites W2963393555 @default.
• W3211882289 cites W2963407932 @default.
• W3211882289 cites W2963419579 @default.
• W3211882289 cites W2963420272 @default.
• W3211882289 cites W2963598268 @default.
• W3211882289 cites W2963717490 @default.
• W3211882289 cites W2963749936 @default.
• W3211882289 cites W2971044234 @default.
• W3211882289 cites W2990500698 @default.
• W3211882289 cites W2994971263 @default.
• W3211882289 cites W2995627237 @default.
• W3211882289 cites W2997593239 @default.
• W3211882289 cites W3000266068 @default.
• W3211882289 cites W3004808386 @default.
• W3211882289 cites W3034676907 @default.
• W3211882289 cites W3034781633 @default.
• W3211882289 cites W3034978746 @default.
• W3211882289 cites W3035058308 @default.
• W3211882289 cites W3035524453 @default.
• W3211882289 cites W3095121901 @default.
• W3211882289 cites W3102696055 @default.
• W3211882289 cites W3104851875 @default.
• W3211882289 cites W3106485021 @default.
• W3211882289 cites W3106539090 @default.
• W3211882289 cites W3108313333 @default.
• W3211882289 cites W3129110783 @default.
• W3211882289 cites W3135958856 @default.
• W3211882289 cites W3151348293 @default.
• W3211882289 cites W3154192710 @default.
• W3211882289 cites W3161430654 @default.
• W3211882289 cites W3163880376 @default.
• W3211882289 cites W3164834180 @default.
• W3211882289 cites W3167945274 @default.
• W3211882289 cites W3168932071 @default.
• W3211882289 cites W3172591507 @default.
• W3211882289 cites W3173143063 @default.
• W3211882289 cites W3192404838 @default.
• W3211882289 cites W3203497602 @default.
• W3211882289 cites W343636949 @default.
• W3211882289 cites W56385144 @default.
• W3211882289 hasPublicationYear "2021" @default.
• W3211882289 type Work @default.
• W3211882289 sameAs 3211882289 @default.
• W3211882289 citedByCount "0" @default.
• W3211882289 crossrefType "posted-content" @default.

• next