BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Rui Zhu (City University of London) DTSTART:20240227T140000Z DTEND:20240227T145000Z DTSTAMP:20260423T024617Z UID:UEA_mth/3 DESCRIPTION:Title: Locally-enriched cross-reconstruction for few-shot fine-grained image clas sification\nby Rui Zhu (City University of London) as part of Fluids a nd Structures Seminar @ UEA\n\nLecture held in TPSC 2.05A.\n\nAbstract\nFe w-shot fine-grained image classification has attracted considerable attent ion in recent years for its realistic setting to imitate how humans conduc t recognition tasks. Metric-based few-shot classifiers have achieved high accuracies. However\, their metric function usually requires two arguments of vectors\, while transforming or reshaping three-dimensional feature ma ps to vectors can result in loss of spatial information. Image reconstruct ion is thus involved to retain more appearance details: the test images ar e reconstructed by different classes and then classified to the one with t he smallest reconstruction error. However\, discriminative local informati on\, vital to distinguish sub-categories in fine-grained images with high similarities\, is not well elaborated when only the base features from a u sual embedding module are adopted for reconstruction. Hence\, we propose t he novel local content-enriched cross-reconstruction network (LCCRN) for f ew-shot fine-grained classification. In LCCRN\, we design two new modules: the local content-enriched module (LCEM) to learn the discriminative loca l features\, and the cross-reconstruction module (CRM) to fully engage the local features with the appearance details obtained from a separate embed ding module. The classification score is calculated based on the weighted sum of reconstruction errors of the cross-reconstruction tasks\, with weig hts learnt from the training process. Extensive experiments on four fine-g rained datasets showcase the superior classification performance of LCCRN compared with the state-of-the-art few-shot classification methods. Codes are available at:https://github.com/lutsong/LCCRN.\n LOCATION:https://researchseminars.org/talk/UEA_mth/3/ END:VEVENT END:VCALENDAR