title: InSAI Lab Profile
emoji: 🏛️
colorFrom: indigo
colorTo: blue
sdk: static
pinned: false
Integrated Sensing and Assistive Intelligence (InSAI) Lab
School of AI and Robotics, Hunan University (HNU)
"Building a utopia where integrated sensing systems empower and enhance human capabilities, especially benefiting people with disabilities."
🎯 About InSAI Lab
The InSAI (pronounced like "Insight") Lab operates within the School of AI and Robotics at Hunan University. Our mission is to merge interdisciplinary fields — including computer vision, robotics, intelligent transportation systems, assistive technologies, and human-computer interaction (HCI) — to develop deployable Assistive Intelligence.
Core Research Directions
- 🌐 Multimodal Perception & Scene Understanding: Vision-Language Modeling, 2D/3D Vision, and sensor fusion in complex, adverse real-world environments.
- 🚗 Autonomous Systems & Panoramic Vision: Semantic Mapping, Panoramic Sensing, Distortion-aware Transformers, and Bird’s-Eye-View (BEV) perception.
- 🤖 Embodied Intelligence: Perception, navigation, and decision-making linking foundation models with robotic action, including Sim2Real adaptation.
- ♿ Assistive Technologies: Wearable assistive systems, navigation aids, and accessible interfaces designed around human capabilities.
- 🧠 Brain-Computer Interface & Document Understanding: Cognitive state sensing, assistive BCI, and intelligent layout/reading assistance.
🌟 Featured Collections & Open-Source Suites
1. 📚 Collection: DeLiVER & CMNeXt (CVPR 2023)
Delivering Arbitrary-Modal Semantic Segmentation
A unified benchmark and cross-modal suite scaling dynamically across 1 to 81 modalities (Depth, LiDAR, Events, Multi-Views, RGB) under 4 severe weather conditions and 5 sensor failure corner cases.
| Resource | Hugging Face Repository | Description |
|---|---|---|
| 📦 Dataset | InSAI-Lab/DELIVER |
Full multimodal dataset package (12.22 GB) |
| 🤖 Model Zoo | InSAI-Lab/CMNeXt |
37 trained checkpoints covering DELIVER, KITTI-360, MFNet, NYU Depth V2, UrbanLF, MCubeS, and SegFormer/Swin backbones |
| 📄 Paper | arXiv:2303.01480 | CVPR 2023 Paper |
# Quick download via Hugging Face CLI
hf download InSAI-Lab/DELIVER DELIVER.tar.gz --type dataset --local-dir ./data
hf download InSAI-Lab/CMNeXt --local-dir ./checkpoints
2. 📚 Collection: Trans4PASS & SynPASS (CVPR 2022)
Transformers for Panoramic Semantic Segmentation & Domain Adaptation
Distortion-aware Vision Transformers tailored for 360° panoramic cameras and omnidirectional scene understanding, featuring the 9,080-frame SynPASS benchmark.
| Resource | Hugging Face Repository | Description |
|---|---|---|
| 📦 Dataset | InSAI-Lab/SynPASS |
9,080 high-res panoramic semantic segmentation benchmark (4.79 GB) |
| 🤖 Model Zoo | InSAI-Lab/Trans4PASS |
40 checkpoints across Cityscapes, Stanford2D3D, Structured3D, and MPA domain adaptation snapshots |
| 📄 Papers | arXiv:2203.01452 / arXiv:2207.11860 | CVPR 2022 & arXiv Preprints |
# Quick download via Hugging Face CLI
hf download InSAI-Lab/SynPASS SynPASS.tar.gz --type dataset --local-dir ./datasets
hf download InSAI-Lab/Trans4PASS --local-dir ./workdirs
🔬 Recent Selected Publications
- IEEE T-PAMI 2026: OneBEV++: Towards Unifying Bird’s-Eye-View Semantic Mapping with Panoramas
- CVPR 2026: More than the Sum: Panorama-Language Models for Adverse Omni-Scenes
- ICML 2026: Position: Assistive Agents Need Accessibility Alignment
- BMVC 2026: X²Localizer: Cross-grained Alignment for Progressive Cross-view Video Geo-localization
- CVPR 2023: Delivering Arbitrary-Modal Semantic Segmentation (DeLiVER & CMNeXt)
- CVPR 2022: Bending Reality: Distortion-aware Transformers for Adapting to Panoramic Semantic Segmentation (Trans4PASS)
🤝 Connect & Collaborate
- 🌐 Official Website: https://insailab.org
- 💻 GitHub Organization: https://github.com/InSAI-Lab
- 📬 Recruitment & Inquiries: jiamingzhang@hnu.edu.cn