BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-LIC-LOCATION:Asia/Tokyo
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
DTSTART:18871231T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260817T171530Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241203T135600
DTEND;TZID=Asia/Tokyo:20241203T141000
UID:siggraphasia_SIGGRAPH Asia 2024_sess104_papers_905@linklings.com
SUMMARY:MVImgNet2.0: A Larger-scale Dataset of Multi-view Images
DESCRIPTION:Xiaoguang Han (SSE, The Chinese University of Hong Kong, Shenz
 hen; FNii, The Chinese University of Hong Kong, Shenzhen); Yushuang Wu, Lu
 yue Shi, Haolin Liu, Hongjie Liao, and Lingteng Qiu (FNii, The Chinese Uni
 versity of Hong Kong, Shenzhen; SSE, The Chinese University of Hong Kong, 
 Shenzhen); Weihao Yuan, Xiaodong Gu, and Zilong Dong (Alibaba); and Shugua
 ng Cui (SSE, The Chinese University of Hong Kong, Shenzhen; FNii, The Chin
 ese University of Hong Kong, Shenzhen)\n\nMVImgNet is a large-scale datase
 t that contains multi-view images of ~220k real-world objects in 238 class
 es. As a counterpart of ImageNet, it introduces 3D visual signals via mult
 i-view shooting, making a soft bridge between 2D and 3D vision. This paper
  constructs the MVImgNet2.0 dataset that expands MVImgNet into a total of 
 ~520k objects and 515 categories, which derives a 3D dataset with a larger
  scale that is more comparable to ones in the 2D domain. In addition to th
 e expanded dataset scale and category range, MVImgNet2.0 is of a higher qu
 ality than MVImgNet owing to four new features: (i) most shoots capture 36
 0-degree views of the objects, which can support the learning of object re
 construction with completeness; (ii) the segmentation manner is advanced t
 o produce foreground object masks of higher accuracy; (iii) a more powerfu
 l structure-from-motion method is adopted to derive the camera pose for ea
 ch frame of a lower estimation error; (iv) higher-quality dense point clou
 ds are reconstructed via advanced methods for objects captured in 360-degr
 ee views, which can serve for downstream applications. Extensive experimen
 ts confirm the value of the proposed MVImgNet2.0 in boosting the performan
 ce of large 3D reconstruction models. MVImgNet2.0 will be public at luyues
 .github.io/mvimgnet2, including multi-view images of all 520k objects, the
  reconstructed high-quality point clouds, and data annotation codes, hopin
 g to inspire the broader vision community.\n\nRegistration Category: Full 
 Access, Full Access Supporter\n\nLanguage Format: English Language\n\nSess
 ion Chair: Bernhard Kerbl (Technical University of Vienna)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_905&sess=sess104
END:VEVENT
END:VCALENDAR