Drag a node to move it. Click or press Enter to open. With a node focused, use Alt + arrow keys to move it; press 0 to reset the map.
Research updates and publications.
Shared the first public version of Vis4GS, our visual analytics tool for inspecting 3D Gaussian Splatting reconstructions.
Read the preprintOur paper on visual analytics for 3D Gaussian Splatting reconstruction was accepted as a short paper at IEEE VIS 2026.
Read the paperResearch, everyday tools, and things built out of curiosity.
Research · IEEE VIS 2026 short paper
A visual analytics tool for inspecting 3D Gaussian Splatting. Linked views connect scene artifacts with Gaussian properties, training history, and densification genealogy.
Explore the research
Developer tool · Rust / Ratatui
A keyboard-driven terminal workspace for projects, persistent AI agent sessions, and system status. Its orbital map turns workspaces into planets and agent sessions into moons.
Explore Waystation
I2P team project · C++ / Allegro
A game project inspired by Age of War, built with C++ and Allegro. Includes stage selection, a shop, and a scoreboard.
Explore the gameAI coursework · Prototype
A course project exploring chest X-ray analysis with computer vision and AI agents. The interface shows detection overlays, generated analysis, and the dialogue between analysis and verification agents.
Explore MedAgentCVSoftware Studio coursework · Flutter
A Flutter and Dart course project designed to turn uploaded syllabi into schedules, with calendar and task management views.
Explore the projectMore experiments and coursework on GitHub.
Browse all repositories ↗IEEE VIS 2026 · Short paper
A Visual Analytic Tool for
3D Gaussian Splatting Reconstruction
Research playground · Real reconstruction outputs
Explore the same view at iterations 500 and 30,000. Look at the lower center: the later reconstruction reveals dark, elongated floater artifacts.
Drag across the image or use the slider to compare. Use “Look closer” to inspect the lower center. The slider also works with your keyboard.
These are two recorded outputs; the divider reveals them without generating intermediate training states.
A final image shows the artifact. Vis4GS links it to individual Gaussians, their properties, and their training history. Explore the research in the paper ↗
Gaussian genealogy · Bonsai
This nine-node family comes from the recorded training log. Select a Gaussian to trace its parents, see when it was created, and check whether it was pruned.
Solid node: not pruned in the recorded snapshot.Dashed node: pruned.
Highlighting follows the selected node back to the root. On a small screen, scroll the tree sideways or choose a node from the menu.
The family begins with Gaussian #3125 at iteration 0. It has six clone children; one of them, #903358, has two further clone children. Enable JavaScript to inspect the tree.
A separate family from the Bonsai scene, showing clone relationships in the iteration-30,000 snapshot. This tree does not identify the Gaussians behind the floater comparison above. Download the recorded family ↗
3D reconstructions can contain blur, floaters, and needle-like artifacts. Vis4GS helps inspect the Gaussians behind those artifacts and trace their evolution during training.
The tool links scene inspection, diagnostic scores, property timelines, and densification genealogy in one visual analysis workflow.
First-authored research, accepted at IEEE VIS 2026.
Explore my research interestsKai-Yuan Lin on papers.
I’m a first-year master’s student in ISA at National Tsing Hua University.
I work in VM Lab, advised by Dr. Shih-Hsuan Hung (Kevin). My interests are in visual analytics and tools that help researchers understand complex computational processes.
B.S. studies in IPTH.
Continued my undergraduate studies at NTHU.
Our short paper on visual analytics for 3D Gaussian Splatting reconstruction.
Shared the first public version on 25 June.
National Tsing Hua University · VM Lab.
Attended the conference and presented Vis4GS.
My interests center on explainable AI (XAI) and visualization.
I’m curious about how visualizations can make AI and complex information easier to explore and understand.
I’m still exploring directions and enjoy exchanging ideas across topics and disciplines.
Exchange ideasPour-over coffee.
Games.
A good workout.
I placed second in a local pour-over coffee competition in Hsinchu.
20 g of coffee · 300 g of water (1:15).
The weights below are cumulative water totals.
Start with 35 g of water to bloom the coffee.
Pour quickly around the coffee bed until the total reaches 80 g.
Make one slow circle, bringing the total to 180 g.
Make one more slow circle, finishing at 300 g.
These are a few things I enjoy outside research. If we meet, coffee is a good place to start.
Say helloI’m interested in research work and collaborations in explainable AI and visualization.
kekn521ISA, National Tsing Hua University
VM Lab