Computer Vision Engineer (Sports Video Analytics)

We are looking for a Computer Vision Engineer to join one of our international clients on an outstaff basis.

Core Responsibilities
Video Algorithmic Pipelines: Design and stitch together custom video processing workflows (Detection Multi-Object Tracking Action Recognition) directly in memory (GPU/CUDA level) to minimize latency.
Camera Calibration & Homography: Develop robust algorithms for automatic pitch/field line detection, lens distortion compensation, and 2D-to-3D coordinate mapping (homography estimation).
Advanced Multi-Object Tracking (MOT): Optimize tracking algorithms (Kalman filters, deep embedding matching) to handle dense crowds, severe occlusions, and ID switches under fast-paced sports dynamics.
Custom Model Adaptation: Fine-tune and structurally modify state-of-the-art CV architectures (YOLO-style detectors, Vision Transformers) specifically for sports domains and low-resolution/far-angle edge cases.
Technical Requirements
CV Experience: 2+ years of production-proven commercial experience in Computer Vision, with a heavy focus on Video Analytics.
Domain Expertise: Solid, demonstrable experience with sports video data, player tracking, or highly dynamic multi-agent scenes.
Hard Skills:Exceptional Python and deep understanding of PyTorch or TensorFlow.
Strong foundation in classical computer vision geometry (projective geometry, epipolar geometry, camera matrices, OpenCV).
Hands-on experience with modern tracking frameworks (e.g., ByteTrack, OC-SORT) or writing proprietary tracking logic.
Experience optimizing models for inference (TensorRT, ONNX, Triton Inference Server).

Engineering Mindset: Ability to profile vision algorithms, identify execution bottlenecks, and optimize for throughput/latency without losing precision.