Yan Shu

Current VLMs lack fine-grained spatial perception. Introducing TerraScope [CVPR 2026]: a pixel-grounded reasoning framework for Earth Observation that interleaves segmentation masks with chain-of-thought for precise geospatial analysis. 🌐 shuyansy.github.io/terrascope/ #CVPR2026 #VLM

arxiv: arxiv.org/pdf/2406.04264 && arxiv.org/pdf/2409.14485 code: github.com/JUNJIE99/MLVU && github.com/VectorSpaceLab…

Excited to see our two papers about long video understanding has been accepted by CVPR 2025. (1) MLVU: Benchmarking Multi-task Long Video Understanding (2) Towards Vision Language Models For Extra-Long Video Understanding #CVPR2025 #CVPR #LLMs