Skip to content
View shjo-april's full-sized avatar
48 followers · 1 following

Achievements

Achievement: Pair Extraordinairex2Achievement: Starstruckx2Achievement: Arctic Code Vault Contributor

Achievements

Achievement: Pair Extraordinairex2Achievement: Starstruckx2Achievement: Arctic Code Vault Contributor

Organizations

@SNU-AIBL

Block or report shjo-april

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
shjo-april/README.md

Hi, I'm Sanghyun 👋

I'm a Principal AI Researcher at OGQ and an Affiliated Researcher at SNU AIBL Lab, working with Prof. Kyungsu Kim. I like finding out what large foundation models already know, and turning that into data-efficient, training-free vision systems that actually ship.

🏠 Homepage · 📄 CV · Google Scholar · OpenReview · ORCID · LinkedIn · ✉️ Email

  • 🔭 Research: weakly-supervised segmentation, multimodal alignment, controllable generation
  • 🧪 Recent obsession: your diffusion model is secretly a lot of things (edge detector included)
  • 🏭 At OGQ since 2017 (9+ years), shipping vision AI used by NAVER, Samsung, JTBC, K-water, and more
  • 📝 Reviewer for NeurIPS, ICLR, CVPR, ICCV, ECCV, ICML, MICCAI, AAAI and IJCV
  • 💬 Always happy to chat about label-efficient learning: shjo.april@gmail.com

🔥 News

  • 2026.06  Two papers accepted to ECCV 2026: ISAC, EraseLoRA
  • 2026.06  GeneRAG accepted to MICCAI 2026
  • 2026.05  ACE accepted to ICML 2026
  • 2026.02  TRACE selected as an Oral at ICLR 2026 ⭐ (Top 1.13%, 223 of 19,735)
  • 2025.06  Two papers accepted to ICCV 2025: COIN, ELECT
  • 2025.06  Named an Outstanding Reviewer at CVPR 2025 🏆 (Top 5.6%)
Earlier news
  • 2024.07  Two papers accepted to ECCV 2024: DHR, TTD
  • 2023.07  MARS accepted to ICCV 2023
  • 2021.01  Puzzle-CAM accepted to ICIP 2021

📚 Publications

Paper titles link to the project page when one exists, otherwise to the GitHub repository. 🟡 marks my leading roles (First / Co-First* / Co-Corresponding†), ⚪ marks co-authored work.

Venue Paper Role Links
ECCV 2026 ISAC: Training-Free Instance-to-Semantic Attention Control for Multi-Instance Generation 🟡 Co-First & Co-Corresponding arXiv · Code
ECCV 2026 EraseLoRA: MLLM-Driven Foreground Exclusion and Background Subtype Aggregation for Dataset-Free Object Removal 🟡 Co-First & Co-Corresponding arXiv · Code
MICCAI 2026 GeneRAG: A Retrieval-Augmented Framework for Spatially Resolved Gene Expression Prediction ⚪ Co-Author Code
ICLR 2026 ⭐ Oral (Top 1.13%) TRACE: Your Diffusion Model is Secretly an Instance Edge Detector 🟡 Co-First arXiv · OpenReview · Code
ICML 2026 ACE: On the Collapse of Generative Paths: A Criterion and Correction for Diffusion Steering ⚪ Co-Author arXiv · OpenReview · Code
ICCV 2025 COIN: Confidence Score-Guided Distillation for Annotation-Free Cell Segmentation 🟡 Co-First arXiv · Code
ICCV 2025 ELECT: Early Timestep Zero-Shot Candidate Selection for Instruction-Guided Image Editing ⚪ Co-Author arXiv · Code
ECCV 2024 DHR: Dual Features-Driven Hierarchical Rebalancing in Inter- and Intra-Class Regions for Weakly-Supervised Semantic Segmentation 🟡 First Author arXiv · Code
ECCV 2024 TTD: Text-Tag Self-Distillation Enhancing Image-Text Alignment in CLIP to Alleviate Single Tag Bias 🟡 Co-First arXiv · Code
ICCV 2023 MARS: Model-agnostic Biased Object Removal without Additional Supervision for Weakly-Supervised Semantic Segmentation 🟡 First Author arXiv · Code
ICIP 2021 Puzzle-CAM: Improved Localization via Matching Partial and Full Features 🟡 First Author arXiv · Code
Under review & preprints
Status Paper Role Links
Under review Chain-of-Prompts: One Click per Cell Type Suffices: Training-free Group Interaction for Cell Instance Segmentation 🟡 First Author arXiv · Code
Under review PatchGate: Your VLM Already Sees the Objects It Forgets to Mention 🟡 Co-Corresponding arXiv TBD
Under review LiftConsensus: Lifting 2D Change Evidence into 3D via Multi-View Spatial Consensus for Retail Change Detection ⚪ Co-Author arXiv TBD
Preprint RSEPM: RecurSeed and EdgePredictMix: Pseudo-Label Refinement Learning for Weakly Supervised Semantic Segmentation 🟡 First Author arXiv · Code

🏭 Research that ships

At OGQ I lead AI R&D that runs in production, not just in papers:

  • 🏷️ Content tagging for NAVER, Samsung and Yonhap News: tens of millions of images auto-tagged per day across 10 languages
  • 🛡️ Content moderation screening 59M+ creator assets for the OGQ markets (NAVER, SOOP, Chat+)
  • 🎬 Generative AI for broadcast with JTBC: AI virtual product placement, cutting post-production cost by ~35%
  • 🚁 Public-safety video analytics for K-water and the Korean National Police Agency: 42 nationwide on-premise sites, 5 KISA certifications
  • 📟 Edge AI with EYENIX: NPU model compression for real-time inference on camera SoCs

Pinned Loading

  1. Chain-of-Prompts Chain-of-Prompts Public

    [Under Review] Chain-of-Prompts (CoP): One Click per Cell Type Suffices, Training-free Group Interaction for Cell Instance Segmentation

    Jupyter Notebook 9

  2. TRACE TRACE Public

    [ICLR 2026 Oral] TRACE: Your Diffusion Model Is Secretly an Instance Edge Detector

    18

  3. COIN COIN Public

    [ICCV 2025] COIN: Confidence Score-Guided Distillation for Annotation-Free Cell Segmentation

    Python 10 3

  4. DHR DHR Public

    [ECCV 2024] DHR: Dual Features-Driven Hierarchical Rebalancing in Inter- and Intra-Class Regions for Weakly-Supervised Semantic Segmentation

    Python 18 2

  5. TTD TTD Public

    [ECCV 2024] TTD: Text-Tag Self-Distillation Enhancing Image-Text Alignment in CLIP to Alleviate Single Tag Bias

    Python 12 2

  6. PuzzleCAM PuzzleCAM Public

    [ICIP 2021] Puzzle-CAM: Improved localization via matching partial and full features.

    Python 180 48