Sebastian Cavada is a computer vision researcher working on 3D scene reconstruction and generation, geometry-grounded foundation models, world models, and Gaussian Splatting. His research focuses on perception systems that reconstruct, ground, and reason about dynamic 3D scenes beyond point clouds and pixels.
After completing an M.Res. in Computer Vision at MBZUAI under Prof. Ian Reid, Sebastian has conducted research with Astra-Vision at Inria Paris, CVI2 at the University of Luxembourg, FBK E3DA, and CovisionLab. His work spans 3D foundation models, camera-aware reconstruction, CAD-oriented multimodal agents, TinyML vision pipelines, physics-grounded benchmarks, and reliable visual mapping for large environments.
News
- 2026: Training-Free Fine-Grained Semantic Segmentations in Low Data Regimes: A FungiTastic Baseline accepted at the 13th Workshop on Fine-Grained Visual Categorization, CVPR 2026. Paper
- 2026: Started as a Research Intern at CovisionLab, working on diffusion-driven synthesis and robustness evaluation for perception models.
- 2025: Collaborated with Inria Paris on NewtPhys, a physics-grounded benchmark for evaluating foundation models on Newtonian reasoning.
- 2025: CAD-Assistant: Tool-Augmented VLLMs as Generic CAD Task Solvers accepted at ICCV 2025. Paper
- 2025: Graduated from MBZUAI with a Master of Research in Computer Vision.
- 2025: All Languages Matter: Evaluating LMMs on Culturally Diverse 100 Languages accepted at CVPR 2025. Paper
- 2024: Does Complexity Pay Off? received Best Paper Award at HEALTHINFO 2024. Paper
- 2023: Graduated from the Free University of Bozen-Bolzano with a B.Sc. in Computer Science.
