James Hun Kim
I am an AI Researcher & Engineer in the Visual AI Group at Samsung Electronics (MX), and an M.S. student in the Interdisciplinary Program in Artificial Intelligence (IPAI) at Seoul National University, advised by Prof. Se Young Chun (ICL).
My research focuses on differentiable rendering, generative models, and 3D vision, with a particular emphasis on optimization-driven systems that bridge research with large-scale production. At Samsung, I led the end-to-end development of a range of AI-powered camera features, from model design to commercialization, deploying real-time vision systems across smartphones, tablets, and foldable devices under strict latency, memory, and power constraints. I am particularly interested in developing efficient and scalable AI systems that translate advanced research into production.
I am a Korean-American. I publish under my Korean name, Junghun James Kim, and also go by my English name, James Hun Kim.
News
- Jul 2026 Selected as a representative student of Seoul National University to present HUG3D at the 2026 Korea AI Innovation Talent Connect (COEX, July 29–31). 🇰🇷
- Jul 2026 HyFL-CLIP paper accepted at ECCV 2026! 🎉
- Jun 2026 Won 3rd place with DiffBMP at the SNU IPAI Research Exchange Meeting. 🏆
- Jun 2026 Human Interaction-Aware 3D Reconstruction paper presented at CVPR 2026 as a Highlight. 🏆
- Jun 2026 Uncertainty-guided Compositional Alignment paper presented at CVPR 2026 as a Highlight. 🏆
- Jun 2026 Presented DiffBMP at CVPR 2026.
- Mar 2025 Started my M.S. in the Interdisciplinary Program in Artificial Intelligence (IPAI) at Seoul National University.
- Mar 2024 Filed a patent on increasing the resolution of digital bokeh images (US20250272806, EP4604568A4).
- Mar 2023 Filed a patent on an adaptive camera pipeline switching between stereo and AI-based bokeh (US2024/0314429, KR10-2794879).
- Nov 2022 Patent on category-aware visual item retrieval granted (US11500918B2, KR10-2734142).
- Feb 2022 Patent on multi-camera video bokeh granted (US11258962B2, KR10-2835449).
- Jan 2019 Joined the Visual AI Group at Samsung MX — leading end-to-end development of Portrait Video, deployed from the Galaxy S10 5G to the Galaxy S25.
- Nov 2018 One-Shot Item Search with Multimodal Data preprint released on arXiv.
- Aug 2017 Moved to the Visual S/W R&D Group at Samsung MX to work on Bixby Vision, an AI-based visual search system.
- Jul 2016 Joined Samsung Electronics (MX).
- May 2015 Selected for the Science & Engineering Startup Nurturing Program, a national grant by NIPA (National IT Industry Promotion Agency), and awarded ₩50,000,000 in funding. 🏆
- May 2014 Selected for the Creative Challenge SW R&D Program, a national R&D grant by NIPA (National IT Industry Promotion Agency), and awarded ₩100,000,000 in funding. 🏆
Selected Publications
* denotes equal contribution. See all publications →
2026
HyFL-CLIP: Hyperbolic Fine-Tuning of CLIP for Robust Long-Context Understanding
European Conference on Computer Vision (ECCV), 2026
A hyperbolic fine-tuning framework that distills the well-established image-text alignment of Euclidean CLIP into hyperbolic space via cross-manifold similarity distillation.
Human Interaction-Aware 3D Reconstruction from a Single Image
Conference on Computer Vision and Pattern Recognition (CVPR), 2026 (Highlight)
A holistic framework for reconstructing physically plausible, high-fidelity textured 3D humans from a single image, explicitly modeling group- and instance-level cues to handle perspective distortion, occlusion, and inter-human interactions.
DiffBMP: Differentiable Rendering with Bitmap Primitives
Conference on Computer Vision and Pattern Recognition (CVPR), 2026
A scalable differentiable renderer for bitmap primitives that optimizes thousands of elements via a highly parallelized custom CUDA pipeline, enabling practical image/video composition, layered export, and artist-friendly creative workflows.
Conference on Computer Vision and Pattern Recognition (CVPR), 2026 (Highlight)
An uncertainty-guided hyperbolic vision-language framework that models part-to-whole semantic representativeness via adaptive uncertainty, improving hierarchical compositional understanding and performance on zero-shot classification, retrieval, and multi-label classification.
Experience
-
Samsung Electronics (MX) — Visual AI Group
Led the end-to-end development and commercialization of Portrait Video, a real-time AI video bokeh system supporting humans, pets, and objects across single, dual, and ToF camera configurations. Deployed across Galaxy product lines from the Galaxy S10 5G to the Galaxy S25. Developed core modules including detection, depth estimation, segmentation, tracking, and visual effects, and built MLOps pipelines for large-scale debugging and evaluation.
-
Samsung Electronics (MX) — Visual S/W R&D Group
Worked on Bixby Vision, an AI-based visual search system. Managed and scaled production systems for Shopping, OCR, and Landmark services, deployed from the Galaxy S8 to the Galaxy Note 10, and established user-feedback-driven evaluation pipelines.
-
Samsung Electronics (MX) — SE Team
Worked on software configuration management (SCM). Automated the firmware release pipeline, reducing required manpower from about 30 engineers to 3.
Patents
-
Electronic device and method for increasing resolution of digital bokeh image
-
Electronic device and method for generating image
-
Electronic device for providing information on item based on category of item
-
Electronic device, method, and computer-readable medium for providing bokeh effect in video
Education
-
Seoul National University
-
Hongik University