BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Seoul
X-LIC-LOCATION:Asia/Seoul
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:KST
DTSTART:18871231T000000
DTSTART:19881009T020000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20230103T035307Z
LOCATION:Auditorium\, Level 5\, West Wing
DTSTART;TZID=Asia/Seoul:20221206T100000
DTEND;TZID=Asia/Seoul:20221206T120000
UID:siggraphasia_SIGGRAPH Asia 2022_sess153_papers_290@linklings.com
SUMMARY:NeuralMarker: A Framework for Learning General Marker Corresponden
 ce
DESCRIPTION:Technical Papers\n\nNeuralMarker: A Framework for Learning Gen
 eral Marker Correspondence\n\nHuang, Pan, Pan, Bian, Xu...\n\nWe tackle th
 e problem of estimating correspondences from a general marker, such as a m
 ovie poster, to an image that captures such a marker. Conventionally, this
  problem is addressed by fitting a homography model based on sparse featur
 e matching. However, they are only able to handle plane-like markers and t
 he sparse features do not sufficiently utilize appearance information. In 
 this paper, we propose a novel framework NeuralMarker, training a neural n
 etwork estimating dense marker correspondences under various challenging c
 onditions, such as marker deformation, harsh lighting, etc. Deep learning 
 has presented an excellent performance in correspondence learning once pro
 vided with sufficient training data. However, annotating pixel-wise dense 
 correspondence for training marker correspondence is too expensive. We obs
 erve that the challenges of marker correspondence estimation come from two
  individual aspects: geometry variation and appearance variation. We, ther
 efore, design two components addressing these two challenges in NeuralMark
 er. First, we create a synthetic dataset FlyingMarkers containing marker-i
 mage pairs with ground truth dense correspondences. By training with Flyin
 gMarkers, the neural network is encouraged to capture various marker motio
 ns. Second, we propose the novel Symmetric Epipolar Distance (SED) loss, w
 hich enables learning dense correspondence from posed images. Learning wit
 h the SED loss and the cross-lighting posed images collected by Structure-
 from-Motion (SfM), NeuralMarker is remarkably robust in harsh lighting env
 ironments and avoids the synthetic image bias. Besides, we also propose a 
 novel marker correspondence evaluation method circumstancing annotations o
 n real marker-image pairs and create a new benchmark. We show that NeuralM
 arker significantly outperforms previous methods and enables new interesti
 ng applications, including Augmented Reality (AR) and video editing.\n\nRe
 gistration Category: FULL ACCESS, EXPERIENCE PLUS ACCESS, EXPERIENCE ACCES
 S, TRADE EXHIBITOR\n\nLanguage: ENGLISH\n\nFormat: IN-PERSON
URL:https://sa2022.siggraph.org/en/full-program/?id=papers_290&sess=sess15
 3
END:VEVENT
END:VCALENDAR