SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
SpAItial is building next-generation World Models that combine generative AI, computer vision, and physics-based 3D simulation. The company is moving beyond 2D pixels to create models that natively understand the geometry and physics of the real world, with applications across robotics, AR/VR, gaming, and cinema.
As a Research Scientist focused on 3D reconstruction, you will advance methods for recovering accurate camera poses and 3D geometry from real-world imagery. Your work will span classical multi-view geometry and state-of-the-art learned reconstruction approaches, including structure-from-motion (SfM), bundle adjustment, SLAM, and feed-forward reconstruction methods. The emphasis is on robustness, accuracy, and generalization across diverse real-world datasets.
Key responsibilities include:
- Designing camera pose estimators and 3D reconstruction pipelines
- Building robust SfM and camera tracking systems for various imaging sensors
- Developing bundle adjusters and nonlinear optimizers, including non-perspective camera models
- Integrating and extending state-of-the-art feed-forward reconstructors (VGGT, DA3, Pi3)
- Advancing deep multi-view stereo, learned matching, and monocular depth estimation methods
- Creating evaluation metrics for pose accuracy and reconstruction quality
- Scaling reconstruction methods to large, diverse datasets while maintaining reliability and efficiency
- Collaborating with researchers to integrate reconstruction advances into production systems
You should have a PhD in computer vision with a strong research focus on 3D reconstruction and publications at top venues (CVPR, ICCV, ECCV, NeurIPS). Deep expertise in multi-view geometry (camera models, epipolar geometry, triangulation, PnP) is essential. Hands-on experience with SfM/SLAM systems (COLMAP, ORB-SLAM) and nonlinear solvers (Ceres) is required. Strong Python and PyTorch skills are necessary, along with comfort debugging failure cases on challenging real-world data. Production experience shipping 3D reconstruction systems is a strong plus.
