Aalto University / Computer Vision

Yejun Zhang

I am a third-year PhD student in Computer Science at Aalto University, Finland, supervised by Prof. Juho Kannala and Prof. Esa Rahtu. My research focuses on 3D generation, and visual localization.

I work on geometry in computer vision, with a focus on visual localization and novel view synthesis. I believe AI can learn 3D geometry, potentially even better than traditional methods. I am especially interested in ViTs, GNNs, Transformers, and diffusion models.

Feel free to drop me a message if you would like to know more about my research or just want to have a casual chat.

Selected Research

GeoMix: Descriptor-Free Visual Localization via Global Context and Multi-Detector Training

Yejun Zhang, Xinjue Wang, Zihan Wang, Esa Rahtu, Juho Kannala

ECCV 2026 European Conference on Computer Vision

paper code

GeoMix is a descriptor-free visual localization framework with directional-distance geometry embeddings, global context nodes, and multi-detector training.

SplatGuide: Geometric Priors from 3D Gaussians for Pose-Free Novel View Synthesis

Yejun Zhang*, Zihan Wang*, Xu Ji*, Yihao Wang, Yuxin Hou, Junyuan Fang, Juho-Matti Kilpelainen, Arno Solin, Hamed Rezazadegan Tavakoli, Esa Rahtu, Juho Kannala

2025 Under review

PDF coming soon code coming soon

SplatGuide reuses a single 3D Gaussian scene for geometric conditioning, occlusion-aware view selection, and feature-level guidance to a multi-view diffusion model.

A2-GNN: Angle-Annular GNN for Visual Descriptor-free Camera Relocalization

Yejun Zhang, Shuzhe Wang, Juho Kannala

3DV 2025 International Conference on 3D Vision

paper code

A2-GNN learns robust geometric structural representations with annular neighbor clustering and distance-angle embeddings for descriptor-free camera relocalization.

Skills

Python, C, SQL, PyTorch, OpenCV, LoRA/PEFT, Git, Linux, LaTeX.