Computer Vision / Deep Learning Engineer

We are looking for a Computer Vision Engineer with strong experience in deep learning and applied computer vision systems. 
The role involves working on challenging CV problems in real-world environments. 
We consider Middle and Senior engineers, with scope adapted to experience. 

Responsibilities 

• Develop, train, and optimize deep learning models for: 

- image retrieval

- image matching (keypoint detection and matching) 

- auxiliary perception tasks supporting the main pipeline 

• Evaluate models using quantitative metrics, including: 

- retrieval quality (mAP, Recall@K) 

- matching performance (precision/recall, repeatability) 

- end-to-end system metrics (accuracy, latency, robustness) 

• Optimize models for production deployment using modern toolchains: 

- ONNX / TensorRT / OpenVINO / edge acceleration where applicable 

- model compression techniques (quantization, pruning, distillation) 

- latency, memory, and throughput optimization 

• Work with large-scale visual datasets and descriptor-based representations 

• Collaborate with engineering teams to integrate models into production systems 

Required Qualifications 

• 2–4+ years of experience in Computer Vision / Deep Learning 

• Hands-on experience with keypoint detection and matching models (e.g. SuperPoint, R2D2, DISK, LightGlue, SuperGlue) 

• Experience with image retrieval or metric learning systems 

• Strong understanding of geometric and motion-related computer vision concepts: 

- keypoint detection and description 

- image matching and geometric verification (RANSAC, homography, PnP) 

- pose estimation and refinement techniques (PnP, bundle adjustment, pose graph optimization) 

- optical flow and frame-to-frame tracking 

- vector search / ANN methods for descriptor retrieval 

• Strong Python skills (PyTorch and scientific computing stack) 

• Ability to read and understand inference code in C++ 

Nice to Have 

• Experience with noisy or imperfect real-world datasets 

• Self- supervised or unsupervised learning methods (contrastive learning, homography supervision, etc.) 

• Experience optimizing models for edge deployment (quantization, pruning, distillation) 

• Familiarity with FAISS or similar vector search systems 

• Experience with optimization libraries for geometric problems (bundle adjustment, pose refinement) 

• Understanding of real-time constraints (latency, memory, CPU inference budgets on ARM or low-power x86 systems)