Jinya Sakurai
  • About
  • News
  • Research
  • Publications
  • Contact

Publications

Research in generative modeling, computer vision, and 3D understanding.

RESEARCH OUTPUT

Selected papers and accompanying resources. Links to project pages, code, and papers are shown when publicly available.

2026

PREPRINT2026

Test-Time Scaling for Safe Text-Guided Image Generation via Intermediate Clean Estimates

Jinya Sakurai, Shueicheng Yan, Xun Xu

A weight-preserving test-time defense for text-to-image diffusion models that detects prohibited concepts from intermediate clean image estimates and intervenes through structured low-rank residual optimization in the text-conditioning space.

arXiv Paper

JOURNALTMLR 2026

FairT2I: Latent Variable Guidance for Training-Free Bias Mitigation with LLM-Assisted Bias Detection

Jinya Sakurai, Yuki Koyama, Issei Sato

FairT2I uses large language models to detect potential social bias in text-to-image generation and rebalances sensitive attributes while preserving generation quality.

arXiv Paper Proceedings

PREPRINT2026

OmniDiMM: Bridging 1D, 2D, and 3D with Any-to-Any Multimodal Modeling

Jason Toskov, Oriol Barbany, Rishubh Singh*, Jinya Sakurai, Efe Tarhan, Oğuzhan Fatih Kar, Roman Bachmann, Amir Zadeh, Jesse Allardice, Chuan Li, Carme Torras, Afshin Dehghan, Amir Zamir

An any-to-any multimodal model that bridges 1D, 2D, and 3D representations for generation, retrieval, and object decomposition across text, images, and 3D modalities.

Project Paper

* Equal contribution

2022

CONFERENCEJSAI 2022

Surface Reconstruction from Raw Point Cloud via Energy-Based Models

Jinya Sakurai, Ryutaro Yamauchi, Ryo Furukawa, Tatsushi Matsubayashi

An energy-based approach to learning implicit surface representations from raw point clouds, designed to remain robust in the presence of point noise.

arXiv Paper Proceedings

Project and GitHub buttons can be added to the same link row whenever those resources become public.