Intelligent Media and Cognition Group
清华大学 · 软件学院 · Tsinghua University, School of Software
iMoonLab is a research group at Tsinghua University, School of Software. We study how intelligent systems can understand complex, unstructured media — from images, text, and 3D shapes to brain networks and knowledge — by modeling high-order relationships beyond pairwise connections.
Our research sits at the intersection of hypergraph computation, multimodal media understanding, and knowledge-driven AI. We build both foundational theory and open-source tools that turn research ideas into reusable code.
| 🧠 Hypergraph Learning | 🎯 Object & Scene Understanding | 🌐 Knowledge & LLMs | 🧬 Brain & Medical AI |
|---|---|---|---|
| High-order neural networks, structure learning, efficient inference | Detection, 3D vision, multi-modal representation | RAG, knowledge extraction, LLM reasoning on hypergraphs | fMRI analysis, disease diagnosis, survival prediction |
Documents → structured knowledge. One command.
uv tool install hyperextract he parse paper.pdf -t general/academic_graph -o ./paper_kb/
| Project | Description | Stars |
|---|---|---|
| DeepHypergraph | A PyTorch library for graph and hypergraph deep learning | |
| THU-HyperG | Hypergraph learning toolbox and benchmarks | |
| Hypergraph-DB | Hypergraph database with built-in visualization |
| Project | Description | Venue | Stars |
|---|---|---|---|
| yolov13 | YOLOv13: real-time detection with hypergraph-enhanced adaptive visual perception | arXiv 2025 | |
| Hyper-YOLO | When visual object detection meets hypergraph computation | IEEE TPAMI 2025 | |
| MeshNet | Mesh neural network for 3D shape representation | AAAI 2019 | |
| HGM2R | Hypergraph-based multi-modal representation for open-set 3D object retrieval | IEEE TPAMI 2024 | |
| PVRNet | Point-view relation neural network for 3D shape recognition | AAAI 2019 |
| Project | Description | Venue | Stars |
|---|---|---|---|
| HGNN | Hypergraph Neural Networks — foundational work on hypergraph convolution | AAAI 2019 | |
| DHGNN | Dynamic Hypergraph Neural Networks for dynamic structure learning | IJCAI 2019 | |
| LightHGNN | Distilling hypergraph neural networks into MLPs for 100× faster inference | ICLR 2024 | |
| KHGNN | Kernel Hypergraph Neural Networks | IEEE TPAMI 2025 | |
| SS-HT | Self-supervised hypergraph training via structure-aware learning | IEEE TPAMI 2025 | |
| HGNN-Shield | Defending hypergraph neural networks against high-order structure attack | IEEE TPAMI |
| Project | Description | Venue | Stars |
|---|---|---|---|
| Hyper-RAG | Combating LLM hallucinations with hypergraph-driven retrieval-augmented generation | Nature Communications 2026 | |
| LLM4Hypergraph | Can large language models comprehend hypergraphs beyond graphs? | ICLR 2025 | |
| HIC | Hypergraph Isomorphism Computation | IEEE TPAMI 2024 |
| Project | Description | Venue | Stars |
|---|---|---|---|
| DHGFormer | Dynamic hierarchical graph transformer for disorder brain disease diagnosis | MICCAI 2025 | |
| LHDFormer | Adaptive embedding for long-range high-order dependencies on fMRI | MICCAI 2025 | |
| DIBrain | Exploring dynamic interpretable brain networks via hierarchical graph transformer | PR 2026 | |
| HGST | Hypergraph-based semantic and topological self-supervised learning for brain disease diagnosis | 2025 | |
| PAGHC | Multimodal hypergraph guide learning for non-invasive ccRCC survival prediction | 2025 |
- 🌐 Homepage: moon-lab.tech
- 💻 GitHub: @iMoonLab
- 📧 Contact: see our homepage for lab contact details
Intelligent Media · Structured Knowledge · Cognition