SumerSports

Computer Vision Engineer

• Build and train CV models for sports video: player/ball detection, multi-object tracking, pose/keypoints, event/action recognition, identity association (re-ID).
• Own the experimentation loop: hypotheses → ablations → error analysis → measurable improvements.
• Design and maintain evaluation: task-appropriate metrics (e.g., MOT metrics, keypoint accuracy, event precision/recall), dataset slices, and failure taxonomy.
• Improve data efficiency: augmentations, sampling strategies, handling label noise, weak/self-supervision where helpful.
• Prototype and iterate on modern architectures (e.g., transformer-based detection/tracking, temporal models, multi-task setups).
• Collaborate on dataset + labeling design: formats, schemas, tooling, versioning.
• Help productionize models: packaging, batch/stream inference patterns, throughput/latency tradeoffs, robustness checks.
• Add lightweight quality gates: reproducibility, automated eval, regression detection.