Brian Wei

I'm an incoming Computer Science PhD student at Washington University in St. Louis, where I'll be working in the Multimodal Vision Research Laboratory led by Dr. Nathan Jacobs. I was previously an undergraduate there studying Data Science and Economics, during which I conducted research advised by Dr. Nan Lin and Dr. Jacobs.

Research

I am interested in multimodal models, representation learning, controllable image and video generation, and developing efficient approaches for large-scale models. My work focuses on applications in remote sensing and natural image domains, and I am also eager to explore a broader range of topics.

We propose TerraDiT-Ω, a diffusion transformer for satellite imagery that is promptable with any geospatial primitive (polygons, polylines, bounding boxes, points) grounded in text.

@inproceedings{wei2026terraditomega,
  title     = {TerraDiT-$\Omega$: Unified Spatial Control for Satellite Image
               Synthesis with Any Geospatial Primitive},
  author    = {Wei, Brian and Sastry, Srikumar and Cher, Daniel and Xing, Eric and Jacobs, Nathan},
  booktitle = {European Conference on Computer Vision},
  year      = {2026}
}

We introduce TTE, a geolocation encoder that learns representations by tessellating the Earth's sphere using Voronoi partitioning.

@inproceedings{cher2026tte,
  title     = {Tesselating The Earth},
  author    = {Cher, Daniel and Iqbal, Hamza and Xing, Eric and Wei, Brian and Jacobs, Nathan},
  booktitle = {European Conference on Computer Vision},
  year      = {2026}
}

We propose TerraDiT, a diffusion transformer for satellite image synthesis conditioned on sparse point locations and text — enabling annotation-efficient, spatially precise generation without dense pixel-level maps.

@article{sastry2026terradit,
  title     = {TerraDiT: Point-Conditioned Diffusion Transformer for Satellite Image Synthesis},
  author    = {Sastry, Srikumar and Cher, Daniel and Wei, Brian and Dhakal, Aayush and Khanal, Subash and Gupta, Dev and Jacobs, Nathan},
  journal   = {arXiv},
  year      = {2026}
}

We introduce VectorSynth, a diffusion-based model that generates pixel-accurate satellite imagery from polygonal geographic annotations with semantic attributes.

@inproceedings{cher2026vectorsynth,
  title     = {VectorSynth: Fine-Grained Satellite Image Synthesis with
               Structured Semantics},
  author    = {Cher, Daniel and Wei, Brian and Sastry, Srikumar and Jacobs, Nathan},
  booktitle = {Winter Conference on Applications of Computer Vision},
  year      = {2026}
}

I served as Co-President of the WashU Robotics Club, following my role as software lead on the rover project. Check out the club website for a look at the projects I was fortunate to work on during my time there!