Keeping imagination
is hard.
Explore my research, ideas, experiments, and the things I build.
What’s new on my publication
All papers are accepted by international conferences or journals.
Image Synthesis Poplar: A Scalable Pipeline for Human-Centric Image Dataset Synthesis
A reproducible Specify-Render-Inspect pipeline for constructing customizable human-centric image-text datasets with explicit provenance and auditable quality control.
Video Understanding Self in Space: Benchmarking Self-Awareness and Spatial Cognition in UAV Embodied Intelligence
We introduce SIS-Bench to evaluate self-awareness and spatial cognition across perception, memory, and reasoning in aerial video understanding, together with SIS-Motion, a motion-aware representation that improves embodied perception, memory, and downstream UAV decision-making.
What’s new on my Collection
Projects, datasets, and related resources are open-sourced on GitHub and Hugging Face.
LaTeX Template OneThree arXiv Template
A reusable LaTeX template for arXiv-style papers, technical reports, and project manuscripts, with a custom title block, author formatting, resource links, teaser image, bibliography, and appendix support.
Dataset MIMO Claude Code Traces 1K
A Claude Code-style coding-agent trace dataset with 1,017 complete trajectories generated by MiMo-V2.5-Pro, covering tool use, reasoning fields, debugging, refactoring, shell workflows, and software-engineering tasks.
Dataset FIFA World Cup Games
A full-match FIFA World Cup video metadata dataset with 40 YouTube matches linked to structured match statistics, lineups, event timelines, formations, venues, and prediction polls.
What’s live in my Workshop
Interactive visualizations, live explorers, and hands-on research interfaces.
Gaokao Gaokao Compass 2025
An interactive admissions data platform for China’s 2025 Gaokao, connecting score ranks, enrollment plans, school admission lines, and major admission lines.
CVPR CVPR 2026 Paper Explorer
A bilingual visualization platform for searching, classifying, and reading structured six-dimensional introductions to CVPR 2026 papers.
What’s new on my Blog
Notes on computer vision, natural language processing, multimodal learning, and research practice.
Research Notes LLMs Can't Jump and Abductive Reasoning
A reflection on abductive reasoning, subjective experience, and why language models can extend existing explanations yet still struggle to make genuine scientific conceptual leaps.
Tutorial Using Claude for Microsoft 365 with a Third-Party API Gateway
How to connect the Claude for Microsoft 365 plugin to a third-party LLM gateway using CLIProxyAPI as a local proxy, bypassing CORS and HTTPS restrictions in the Office WebView sandbox.