Computer Vision & 3D Reconstruction Engineer [Remote Sensing x 3D Reconstruction x GIS / Construction DX]

[About Skymatix] Skymatix develops remote sensing technology to address challenges in construction, forestry, infrastructure, and disaster management. Our mission is to create a new society through remote sensing. Our core product, KUMIKI, is a spatial data integration platform that brings together photographs, videos, point clouds, orthophotos, and other data captured by drones, satellites, and 360-degree cameras on a single map. It helps field teams, headquarters, and management understand site conditions and improve surveying, inspection, and measurement workflows. This role supports these applications by developing the technology to generate accurate 3D data from diverse imagery. [Position overview] Develop algorithms and production software to reconstruct accurate 3D geometry from satellite, aerial, and UAV imagery. [Why this role matters now] We are seeking an engineer to advance our satellite imagery reconstruction projects and extend our capabilities to emerging imaging platforms, including 360-degree cameras. You will apply your understanding of computer vision and geometry to develop, improve, and implement 3D reconstruction algorithms. [Key responsibilities] Develop and improve photogrammetric reconstruction pipelines. Your responsibilities will reflect your expertise and project priorities, and may include: - SfM and MVS: Develop and improve feature detection and matching, camera pose estimation, triangulation, depth estimation, and point cloud generation. - Geometric estimation and optimization: Implement bundle adjustment and joint optimization of camera parameters and 3D structure using Ceres Solver or equivalent frameworks. - Satellite reconstruction: Develop methods that account for sensor geometry, viewing direction, georeferencing, and large-scale spatial coverage. - 360-degree imaging: Investigate camera models, calibration, and pose estimation, and develop methods for integrating omnidirectional imagery into reconstruction pipelines. - Quality and performance: Evaluate and improve accuracy, completeness, robustness, and computational efficiency. Depending on your expertise, work on point cloud preprocessing, registration, meshing, and surface reconstruction. - Production integration: Collaborate with research and engineering teams to integrate algorithms into scalable, maintainable 3D reconstruction and geospatial processing systems. [Career opportunities] - Apply multi-view geometry and numerical optimization to software that processes real satellite and aerial imagery. - Contribute across algorithm development, implementation, accuracy evaluation, and performance optimization. - Work in an English-speaking team and expand your experience across imaging platforms, from satellite imagery to 360-degree cameras. [Development environment] C++ / Python; Linux; NVIDIA Docker; GeForce RTX / Tesla GPUs; AWS EC2 GPU instances; VS Code Dev Containers; Git / GitHub; Slack / Google Meet. [Learn more] - KUMIKI spatial data integration platform 　https://smx-kumiki.com/ - Corporate website 　https://skymatix.co.jp/ - Company presentation 　https://speakerdeck.com/skymatix1111/company-profile - Skymatix on note 　https://note.com/skymatix

[Required qualifications] Meet the core requirements below and bring strong knowledge of at least one specialist area. Experience in every specialist area is not required. [Core requirements] - Practical C++ and/or Python implementation experience, with the ability to implement, debug, profile, or extend computer vision, image processing, or 3D reconstruction algorithms. - Understanding of 3D reconstruction fundamentals, including point clouds, coordinate frames, geometric consistency, and factors affecting reconstruction accuracy and completeness. - Ability to read, write, and discuss technical topics in English. English is the primary language used within the team. [At least one specialist area] - Structure-from-Motion (SfM): Feature detection and matching, geometric verification, relative and camera pose estimation, triangulation, and incremental or global reconstruction. - Multi-View Stereo (MVS) and dense reconstruction: Depth estimation, multi-view consistency, depth map fusion, and dense point cloud generation. - Bundle adjustment and nonlinear optimization: Reprojection error minimization, robust loss functions, and joint camera and structure optimization; knowledge of Ceres Solver, g2o, GTSAM, or equivalent frameworks. - Multi-view geometry: Camera models, intrinsic and extrinsic calibration, epipolar geometry, essential and fundamental matrices, coordinate transformations, and geometric constraints.

[Preferred qualifications] - Satellite stereo reconstruction, sensor models such as RPCs, DSMs/DEMs, orthorectification, or large-scale terrain and urban reconstruction. - 360-degree or fisheye camera models and calibration, or geometric estimation and reconstruction using non-pinhole cameras. - Use or extension of COLMAP, OpenMVG, or OpenMVS; or point cloud processing, registration, meshing, and surface reconstruction. - GPU acceleration and parallel computing; software development using Linux, Git, CMake, and automated testing. - Japanese communication skills.

[Who will thrive in this role] - You understand the mathematical and geometric principles behind algorithms and enjoy both implementation and evaluation with real data. - You can identify issues in accuracy, robustness, and processing speed, and use evaluation results to guide improvements. - You are interested in translating research results into software that can be used and maintained over time. - You share technical knowledge and experimental results in English and collaborate effectively with colleagues.