BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Australia/Melbourne
X-LIC-LOCATION:Australia/Melbourne
BEGIN:DAYLIGHT
TZOFFSETFROM:+1000
TZOFFSETTO:+1100
TZNAME:AEDT
DTSTART:19721003T020000
RRULE:FREQ=YEARLY;BYMONTH=4;BYDAY=1SU
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:19721003T020000
TZOFFSETFROM:+1100
TZOFFSETTO:+1000
TZNAME:AEST
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260114T163646Z
LOCATION:Meeting Room C4.8\, Level 4 (Convention Centre)
DTSTART;TZID=Australia/Melbourne:20231212T170000
DTEND;TZID=Australia/Melbourne:20231212T171000
UID:siggraphasia_SIGGRAPH Asia 2023_sess142_papers_644@linklings.com
SUMMARY:FuseSR: Super Resolution for Real-time Rendering through Efficient
  Multi-resolution Fusion
DESCRIPTION:Zhihua Zhong (State Key Lab of CAD&CG, Zhejiang University; Zh
 ejiang University City College); Jingsen Zhu (State Key Lab of CAD&CG, Zhe
 jiang University); Yuxin Dai (Zhejiang A&F University); Chuankun Zheng (St
 ate Key Lab of CAD&CG, Zhejiang University); Guanlin Chen (Zhejiang Univer
 sity City College); Yuchi Huo (Zhejiang Lab; State Key Lab of CAD&CG, Zhej
 iang University); and Hujun Bao and Rui Wang (State Key Lab of CAD&CG, Zhe
 jiang University)\n\nThe workload of real-time rendering is steeply increa
 sing as the demand for high resolution, high refresh rates, and high reali
 sm rises, overwhelming most graphics cards. To mitigate this problem, one 
 of the most popular solutions is to render images at a low resolution to r
 educe rendering overhead, and then manage to accurately upsample the low-r
 esolution rendered image to the target resolution, a.k.a. super-resolution
  techniques. Most existing methods focus on exploiting information from lo
 w-resolution inputs, such as historical frames. The absence of high freque
 ncy details in those LR inputs makes them hard to recover fine details in 
 their high-resolution predictions. In this paper, we propose an efficient 
 and effective super-resolution method that predicts high-quality upsampled
  reconstructions utilizing low-cost high-resolution auxiliary G-Buffers as
  additional input. With LR images and HR G-buffers as input, the network r
 equires to align and fuse features at multi resolution levels. We introduc
 e an efficient and effective H-Net architecture to solve this problem and 
 significantly reduce rendering overhead without noticeable quality deterio
 ration. Experiments show that our method is able to produce temporally con
 sistent reconstructions in $4 \times 4$ and even challenging $8 \times 8$ 
 upsampling cases at 4K resolution with real-time performance, with substan
 tially improved quality and significant performance boost compared to exis
 ting works.\n\nRegistration Category: Full Access\n\nSession Chair: Michae
 l Gharbi (Reve AI, Massachusetts Institute of Technology (MIT))\n\n
URL:https://asia.siggraph.org/2023/full-program?id=papers_644&sess=sess142
END:VEVENT
END:VCALENDAR