Computer Vision Engineer

This is a fully remote position, open to applicants in United States.

📋 Description

• Develop and train computer vision models for sports videos, including player and ball detection, multi-object tracking, pose/keypoint identification, event/action recognition, and identity association (re-ID).

• Take ownership of the experimentation process: formulate hypotheses, conduct ablations, perform error analysis, and implement measurable improvements.

• Create and manage evaluation frameworks: establish task-relevant metrics (such as MOT metrics, keypoint accuracy, and event precision/recall), dataset slices, and failure classifications.

• Enhance data efficiency through augmentations, sampling techniques, addressing label noise, and applying weak/self-supervision as needed.

• Prototype and refine contemporary architectures, including transformer-based detection/tracking, temporal models, and multi-task configurations.

• Collaborate on the design of datasets and labeling processes: including formats, schemas, tools, and version control.

• Assist in the productionization of models: focusing on packaging, batch and stream inference patterns, throughput and latency trade-offs, and robustness assessments.

• Implement lightweight quality control measures: ensuring reproducibility, automated evaluation, and regression detection.

⛳️ Requirements

• Extensive applied experience in computer vision with a focus on hands-on model development (not merely executing existing repositories).

• Proficient in PyTorch: adept at training loops, debugging, and creating data pipelines for vision tasks, as well as understanding DDP fundamentals.

• Familiarity with video computer vision principles: including occlusion, identity changes, temporal consistency, calibration, and domain shifts.

• Strong Python programming skills with a focus on achieving measurable results.

• Preferred (Bonus): Experience in sports video computer vision or related fields (such as multi-agent tracking, pose estimation, or crowded scene analysis).

• Knowledge of video processing tools (like FFmpeg), efficient dataset formats (such as WebDataset/shards), or techniques for streaming/batching to GPUs.

• Experience in MLOps and production environments: including model packaging, continuous integration for training and evaluation, serving (Triton/TorchServe), and monitoring.

🏝️ Benefits

• Comprehensive health insurance coverage.

• Retirement savings plan (401k) with company matching contributions.

• Generous paid holiday schedule totaling 13 days, including the Monday after the Super Bowl.

• Flexible remote working environment.