Responsibilities
- Development and optimization of the Computer Vision pipeline
- Working with video streams and processing low-quality video input
- Implementation of cross-view feature matching and image retrieval
- Building and indexing a georeferenced tile database
- Object georeferencing and spatial analytics
- Optimization of inference latency and performance
- Participation in system architecture design
Must have
- 4+ years of experience in Computer Vision / Deep Learning
- Hands-on experience with geo-localisation, place recognition, or image retrieval
- Feature extraction with foundation models: DINOv2, CLIP, or equivalent
- Keypoint detection and matching: LightGlue, LoFTR, DISK, SuperPoint
- Contrastive / metric learning: InfoNCE, triplet loss, hard negative mining
- Homography estimation and geometric verification
- Experience with satellite imagery and geospatial data
- Coordinate systems, tile indexing, GDAL
- Vector search: FAISS or other ANN libraries
- Python, PyTorch at production level
- Experience with video pipelines
- Linux, Docker
Nice to have
- PhD in Computer Vision, Photogrammetry, Remote Sensing, Geomatics, or Robotics
- Publications in cross-view localisation, visual SLAM, or UAV navigation
- Experience with degraded or compressed video
- TensorRT, ONNX, edge deployment
- CUDA
- Background in defense, aerospace, or robotics
Why should you apply
• Here you will work not only with machine learning models, but also with real production systems and tasks used in a combat environment
• You will have the opportunity to independently build architecture, data processing pipelines and influence key technical decisions, and not just perform individual tasks
• We value a strong engineering approach and people who are able to think systematically, analyze problems and find practical solutions
• The work uses modern research approaches - the team constantly analyzes and implements new research works and current technologies
• Great attention is paid to system performance, optimization of inference, GPU pipeline, latency and scalability
• This is an environment where an engineer really influences the product, technologies and the final result, and does not work within the narrow framework of a separate module