BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260417T190159Z
LOCATION:West Building\, Level 2\, Outside Room 219
DTSTART;TZID=America/Los_Angeles:20250813T090000
DTEND;TZID=America/Los_Angeles:20250813T173000
UID:siggraph_SIGGRAPH 2025_sess438@linklings.com
SUMMARY:Posters: Images, Video & Computer Vision
DESCRIPTION:55. Train Once, Generate Anywhere: Discretization Agnostic Neu
 ral Cellular Automata Using SPH Method\n\nWe introduce SPH‑NCA, a discreti
 zation agnostic neural cellular automata that uses a differentiable SPH me
 thod for perception and a stable training scheme, allowing image and textu
 re synthesis on any grid, resolution, or 3D surface while trained on a fix
 ed-resolution 2D image.\n\n\nHyunsoo Kim and Jinah Park (Korea Advanced In
 stitute of Science and Technology (KAIST))\n---------------------\n54. Tak
 ing Control: Procedural Diffusion Guidance for Architectural Facade Editin
 g\n\nOur training-free method enables photorealistic facade editing by com
 bining hierarchical procedural structure control with diffusion models. St
 arting from a facade image, we reconstruct, edit, and guide generation to 
 produce high-fidelity, photorealistic variations. The method ensures struc
 tural con...\n\n\nAleksander Plocharski (Warsaw University of Technology, 
 IDEAS NCBR); Jan Swidzinski (IDEAS NCBR); and Przemyslaw Musialski (New Je
 rsey Institute of Technology (NJIT))\n---------------------\n52. StructInb
 et: Integrating Explicit Structural Guidance Into Inbetween Frame Generati
 on\n\nStructInbet is an skeleton-based inbetweening system that achieves c
 ontrollable, structure-aware interpolation generation with improved pose c
 larity and motion alignment to user intent, surpassing prior point-based m
 ethods in reducing ambiguity.\n\n\nZhenglin Pan and Haoran Xie (Japan Adva
 nced Institute of Science and Technology)\n---------------------\n46. Pred
 icting Colors in Unpainted Gaps for Anime-Style Illustration\n\nWe introdu
 ce the novel task of predicting flat colors for unintended small regions l
 eft unpainted by flood-fill operations—common in anime-style illustrations
 —and present a U-Net-based method that achieves 62.5% exact-match accuracy
  on professional data, outperforming naïve baselines and...\n\n\nMasahiro 
 Kono (The University of Tokyo); Akinobu Maejima (OLM Digital, Inc.; IMAGIC
 A GROUP Inc.); and Yuki Koyama and Takeo Igarashi (The University of Tokyo
 )\n---------------------\n42. From Style to Identity: AI Pipelines for Vis
 ual and Character Coherence in Film\n\nWe introduce a modular, open-source
  pipeline that combines multiple custom-trained LoRA and ControlNet models
  to disentangle style and identity, enabling fast, visually and narrativel
 y consistent AI-generated short films，validated through two award-winning 
 multi-scene productions.\n\n\nZhiyu Zhang (Brown University, Rhode Island 
 School of Design) and Rui Wang (Independent Researcher)\n-----------------
 ----\n48. Reconstructing Graphic Design Posters via Visual Decomposition a
 nd Semantic Layer Translation\n\nThis work presents a pipeline that conver
 ts rasterized graphic design posters into multi-layered, editable assets. 
 It decomposes elements, addresses layer ordering using a novel Z-index str
 ategy, and shows high accuracy through evaluations of over 24,000 posters.
  User feedback confirms its ability t...\n\n\nVeeramanohar Avudaiappan and
  Ritwik Murali (Amrita Vishwa Vidyapeetham)\n---------------------\n50. Sk
 etch-based Fluid Video Generation Using Motion-Guided Diffusion Models in 
 Still Landscape Images\n\nWe propose a finetuned conditional latent diffus
 ion model for generating motion field from user-provided sketches, which a
 re subsequently integrated into a latent video diffusion model via a motio
 n\nadapter to precisely control the fluid movement.\n\n\nHao Jin and Haora
 n Xie (Japan Advanced Institute of Science and Technology)\n--------------
 -------\n39. Automatic Interpretation of Ancient Egyptian Texts for Educat
 ion and Research\n\nWe present the first open-source system for automatic 
 interpretation of Ancient Egyptian texts, combining OCR, transliteration, 
 and translation into a unified pipeline that supports diverse writing styl
 es and improves accessibility for learners and researchers.\n\n\nMaksim Go
 lyadkin (AIRI, HSE University); Innokentiy Humonen (AIRI); Yanis Plevokas,
  Ekaterina Bureeva, and Ekaterina Alexandrova (HSE University); and Ilya M
 akarov (AIRI, ISP RAS)\n---------------------\n53. Super Resolution for Hu
 mans\n\nWe introduce an architecture-agnostic super-resolution framework t
 hat uses human visual sensitivity to allocate computational resources effi
 ciently, delivering substantial reductions in computational demand without
  perceptible quality loss, as validated by user studies—offering significa
 nt adv...\n\n\nVolodymyr Karpenko, Taimoor Tariq, Jorge Condor, and Piotr 
 Didyk (Università della Svizzera Italiana)\n---------------------\n51. Sti
 cking Information in Plain Sight: Encoding and Detecting Hidden Stickers i
 n the Real World\n\nWe present a pipeline for designing and detecting subt
 le code-conveying patterns that can be\nprinted on transparent sticker pap
 er, then applied to real-world surfaces, rendering the modifications imper
 ceptible to the human eye, but robustly detectable to our model, with spec
 ific emphasis placed on a...\n\n\nChristina Shatford and Szymon Rusinkiewi
 cz (Princeton University)\n---------------------\n43. Full-Color Natural L
 ight Holographic Video Camera\n\nWe present a compact, handheld holographi
 c video camera that captures full-color holograms in real time under natur
 al lighting, making laser-free holography possible. By integrating advance
 d optical components and AI-driven super-resolution, it enables high-quali
 ty holographic content capture, pavin...\n\n\nKihong Choi, Daeyoul Park, a
 nd Keehoon Hong (Electronics and telecommunications research institute)\n-
 --------------------\n47. QRBTF - AI QR Code Generator\n\nQRBTF is an AI Q
 R code generator trained with ControlNet, which can generate scannable QR 
 codes hidden within images based on prompt input.\n\n\nHao Ni (Tongji Univ
 ersity, Latent Cat Inc); Baiyu Chen (Latent Cat Inc); Zhaohan Wang (Commun
 ication University of China, Latent Cat Inc); Zhiyong Chen (Nanjing Univer
 sity, Latent Cat Inc); Wanyi Miao and Lyu Xin (Communication University of
  China); and Nan Cao (Tongji University)\n---------------------\n45. Physi
 cally-Based Compositing of 2D Graphics\n\nWe propose a geometry- and illum
 ination-aware 2d-graphic compositing pipeline. We use meshes generated by 
 off-the-shelf monocular depth estimation methods to warp the 2d-graphic ac
 cording to the surface geometry. Using intrinsic decomposition, we composi
 te the warped graphic onto the albedo and reco...\n\n\nTyrus Tracey, Stefa
 n Diaconu, Sebastian Dille, S. Mahdi H. Miangoleh, and Yağız Aksoy (Simon 
 Fraser University)\n---------------------\n56. Two-Stage Sketch-Based Smok
 e Illustration Generation Using Stream Function\n\nWe propose a two-stage 
 sketch-guided smoke illustration generation framework using stream functio
 n. The input sketch is converted into the stream function through a latent
  diffusion model, which subsequently drives the velocity field generation.
  The velocity field serves as a guidance force to drive...\n\n\nHengyuan C
 hang and Xiaoxuan Xie (Japan Advanced Institute of Science and Technology)
 , Syuhei Sato (Hosei University), and Haoran Xie (Japan Advanced Institute
  of Science and Technology)\n---------------------\n40. Confidence Estimat
 ion of Few-Shot Patch-Based Learning for Anime-Style Colorization\n\nThis 
 study proposes a region-wise confidence estimation method for anime-style 
 line drawing colorization. By comparing local patches in the colorized ima
 ge with training images using normalized cross-correlation, the method hig
 hlights uncertain regions. It improves usability by aiding artists in ide.
 ..\n\n\nYuexiang Ji (The University of Tokyo); Akinobu Maejima (OLM Digita
 l, Inc.; IMAGICA GROUP Inc.); and Yotam Sechayk, Yuki Koyama, and Takeo Ig
 arashi (The University of Tokyo)\n---------------------\n49. SAWNA: Space-
 Aware Text to Image Generation\n\nSAWNA tackles layout-sensitive text-to-i
 mage generation by treating user-specified empty regions as first-class co
 nstraints. Bounding-box masks are blurred and injected as mean-shifted, in
 ert noise into the frozen Stable Diffusion latent, suppressing synthesis i
 nside reserved areas while preserving ...\n\n\nRyugo Morita (Hosei Univers
 ity, EQUES Inc.); Sho Kuno (The University of Tokyo, EQUES Inc.); Ryunosuk
 e Tanaka (EQUES Inc.); Rongzhi Li (The University of Tokyo, EQUES Inc.); a
 nd Hoang Dai Dinh and Issey Sukeda (EQUES Inc.)\n---------------------\n38
 . Assessing Learned Models for Phase-only Hologram Compression\n\nWe evalu
 ate four models using INR and VAE structures for compressing phase-only ho
 lograms. Our findings show that the pretrained VAE struggles with this tas
 k, while SIREN achieves 40% compression with high-quality 3D images (PSNR 
 = 34.54 dB), highlighting the effectiveness of INRs and VAE limitation...\
 n\n\nZicong Peng and Yicheng Zhan (University College London (UCL)), Josef
  Spjut (NVIDIA), and Kaan Akşit (University College London (UCL))\n-------
 --------------\n37. Anime Colorization Using Segment Matching With Candida
 te Colors\n\nA novel method for automatic colorization of anime line drawi
 ngs achieves improved accuracy over state-of-the-art segment matching-base
 d approaches by leveraging semantic segmentation and color shuffling proce
 sses without relying on flow estimation, effectively addressing challenges
  posed by large m...\n\n\nYu Takano (Waseda University); Akinobu Maejima (
 OLM Digital Inc., IMAGICA GROUP Inc.); and Shugo Yamaguchi and Shigeo Mori
 shima (Waseda University)\n---------------------\n41. Emulating Emulsion: 
 A Compact Physically-Based Model for Film Colour\n\nA 30-parameter, physic
 s-based model transforms digital images into authentically scanned film co
 lour. Trained on a single roll of colour-positive film, it matches LUT acc
 uracy without artefacts and exposes interpretable parameters, offering fil
 mmakers a data-light and production-ready solution to re...\n\n\nHyun Jo J
 ang (University of Toronto) and Hakki Karaimer and Michael Brown (AI Cente
 r - Toronto, Samsung Electronics)\n---------------------\n44. G-FED: G-Buf
 fer Guided Frame Extrapolation in Video Diffusion Models\n\nTo generate pr
 eviews with near-final render quality in VFX and enable faster iteration, 
 we propose G-FED, G-Buffer Guided Frame Extrapolation in Video Diffusion M
 odels. G-FED denoises 1spp frames, guided by G-buffer data, to infill mask
 ed forward projections and generate high-quality images that are...\n\n\nP
 edro Pena, Karthik Mohan Kumar, Damian Andrysiak, Kunal Tyagi, and Rama Ha
 rihara (Advanced Micro Devices (AMD))\n\nRegistration Category: Full Confe
 rence, Experience
END:VEVENT
END:VCALENDAR