FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3212821683> ?p ?o ?g. }

• W3212821683 abstract "Recently, a series of algorithms have been explored for GAN compression,which aims to reduce tremendous computational overhead and memory usages whendeploying GANs on resource-constrained edge devices. However, most of theexisting GAN compression work only focuses on how to compress the generator,while fails to take the discriminator into account. In this work, we revisitthe role of discriminator in GAN compression and design a novelgenerator-discriminator cooperative compression scheme for GAN compression,termed GCC. Within GCC, a selective activation discriminator automaticallyselects and activates convolutional channels according to a local capacityconstraint and a global coordination constraint, which help maintain the Nashequilibrium with the lightweight generator during the adversarial training andavoid mode collapse. The original generator and discriminator are alsooptimized from scratch, to play as a teacher model to progressively refine thepruned generator and the selective activation discriminator. A novel onlinecollaborative distillation scheme is designed to take full advantage of theintermediate feature of the teacher generator and discriminator to furtherboost the performance of the lightweight generator. Extensive experiments onvarious GAN-based generation tasks demonstrate the effectiveness andgeneralization of GCC. Among them, GCC contributes to reducing 80%computational costs while maintains comparable performance in image translationtasks. Our code and models are available at url{https://github.com/SJLeo/GCC}." @default.
• W3212821683 created "2021-11-22" @default.
• W3212821683 creator A5000389309 @default.
• W3212821683 creator A5006837984 @default.
• W3212821683 creator A5016080094 @default.
• W3212821683 creator A5017788681 @default.
• W3212821683 creator A5034191292 @default.
• W3212821683 creator A5045089479 @default.
• W3212821683 date "2021-10-27" @default.
• W3212821683 modified "2023-09-25" @default.
• W3212821683 title "Revisiting Discriminator in GAN Compression: A Generator-discriminator Cooperative Compression Scheme." @default.
• W3212821683 cites W1690739335 @default.
• W3212821683 cites W1710476689 @default.
• W3212821683 cites W1821462560 @default.
• W3212821683 cites W1834627138 @default.
• W3212821683 cites W1861492603 @default.
• W3212821683 cites W1901129140 @default.
• W3212821683 cites W1930824406 @default.
• W3212821683 cites W2047920195 @default.
• W3212821683 cites W2108598243 @default.
• W3212821683 cites W2114770744 @default.
• W3212821683 cites W2121927366 @default.
• W3212821683 cites W2124659975 @default.
• W3212821683 cites W2133665775 @default.
• W3212821683 cites W2167215970 @default.
• W3212821683 cites W2173520492 @default.
• W3212821683 cites W2194775991 @default.
• W3212821683 cites W2267635276 @default.
• W3212821683 cites W2279098554 @default.
• W3212821683 cites W2286365479 @default.
• W3212821683 cites W2331128040 @default.
• W3212821683 cites W2340897893 @default.
• W3212821683 cites W2475287302 @default.
• W3212821683 cites W2515385951 @default.
• W3212821683 cites W2585630030 @default.
• W3212821683 cites W2593414223 @default.
• W3212821683 cites W2604763608 @default.
• W3212821683 cites W2605135824 @default.
• W3212821683 cites W2619459889 @default.
• W3212821683 cites W2739748921 @default.
• W3212821683 cites W2766527293 @default.
• W3212821683 cites W2768959015 @default.
• W3212821683 cites W2804078698 @default.
• W3212821683 cites W2807033398 @default.
• W3212821683 cites W2891158090 @default.
• W3212821683 cites W2893749619 @default.
• W3212821683 cites W2951569836 @default.
• W3212821683 cites W2952773607 @default.
• W3212821683 cites W2953246223 @default.
• W3212821683 cites W2962793481 @default.
• W3212821683 cites W2962850830 @default.
• W3212821683 cites W2962851801 @default.
• W3212821683 cites W2962928367 @default.
• W3212821683 cites W2963073614 @default.
• W3212821683 cites W2963163009 @default.
• W3212821683 cites W2963373786 @default.
• W3212821683 cites W2963420272 @default.
• W3212821683 cites W2963446712 @default.
• W3212821683 cites W2963470893 @default.
• W3212821683 cites W2963674932 @default.
• W3212821683 cites W2963981733 @default.
• W3212821683 cites W2964193438 @default.
• W3212821683 cites W2971183298 @default.
• W3212821683 cites W2983484869 @default.
• W3212821683 cites W2997563695 @default.
• W3212821683 cites W3035200909 @default.
• W3212821683 cites W3035204081 @default.
• W3212821683 cites W3094786017 @default.
• W3212821683 cites W3098099022 @default.
• W3212821683 cites W3100565014 @default.
• W3212821683 cites W3148140980 @default.
• W3212821683 cites W3157730339 @default.
• W3212821683 cites W3173562028 @default.
• W3212821683 cites W3173855227 @default.
• W3212821683 cites W3177640626 @default.
• W3212821683 cites W3186169520 @default.
• W3212821683 cites W3195362412 @default.
• W3212821683 hasPublicationYear "2021" @default.
• W3212821683 type Work @default.
• W3212821683 sameAs 3212821683 @default.
• W3212821683 citedByCount "0" @default.
• W3212821683 crossrefType "posted-content" @default.
• W3212821683 hasAuthorship W3212821683A5000389309 @default.
• W3212821683 hasAuthorship W3212821683A5006837984 @default.
• W3212821683 hasAuthorship W3212821683A5016080094 @default.
• W3212821683 hasAuthorship W3212821683A5017788681 @default.
• W3212821683 hasAuthorship W3212821683A5034191292 @default.
• W3212821683 hasAuthorship W3212821683A5045089479 @default.
• W3212821683 hasConcept C113775141 @default.
• W3212821683 hasConcept C121332964 @default.
• W3212821683 hasConcept C159985019 @default.
• W3212821683 hasConcept C163258240 @default.
• W3212821683 hasConcept C180016635 @default.
• W3212821683 hasConcept C192562407 @default.
• W3212821683 hasConcept C2524010 @default.
• W3212821683 hasConcept C2776036281 @default.
• W3212821683 hasConcept C2779803651 @default.
• W3212821683 hasConcept C2780992000 @default.
• W3212821683 hasConcept C33923547 @default.
• W3212821683 hasConcept C41008148 @default.

• next