Senior Computer Vision Engineer

About DevSavant

At DevSavant, we are a trusted technology partner specializing in Software Development, Data Engineering, AI/Machine Learning, Cloud Solutions, Automation Testing, and UI/UX Design. We deliver innovative, high-quality solutions with a focus on excellence and results. Our people are at the heart of everything we do, fostering a culture of growth and well-being. Join us and thrive in a supportive, success-driven environment.

About the Role

We’re looking for a highly skilled Senior Computer Vision Engineer to join our team and drive the development of advanced vision systems across video-based applications. You’ll be working on bleeding-edge technologies in pose estimation, object detection, and motion tracking—optimizing deep learning models and deploying them at scale.

This role is ideal for an engineer who thrives in high-performance environments, understands the complexity of visual inference at scale, and wants to contribute to real-world applications in sports, motion analysis, and other high-frame-rate video contexts.

Key Responsibilities

Pose Estimation & Object Detection

• Build and optimize pose estimation pipelines using HRNet, including multi-person and sports use cases.

• Implement keypoint detection models with multi-stage refinement and heatmap generation.

• Customize and train YOLOv4/X for object detection, optimizing anchor tuning, NMS, and accuracy-speed balance.

• Evaluate and compare top-down vs bottom-up pose estimation approaches.
  

Motion Tracking & Segmentation

• Apply tracking techniques including Kalman Filters, DeepSort, and custom tracking heuristics.

• Analyze temporal coherence in high-FPS sequences (30–120fps).

• Implement segmentation models for semantic and instance-level use cases using both classical and deep learning methods.
  

Model Engineering & Optimization

• Convert models from PyTorch → ONNX → TensorRT, leveraging ONNXRuntime for inference.

• Apply quantization, pruning, batching, and memory optimization techniques to deploy models efficiently at scale.

• Optimize GPU memory usage and runtime throughput for large-scale inference workloads.
  

Video Processing at Scale

• Manage full-frame inference for high-FPS video using tools like FFmpeg, OpenCV, and Decord.

• Implement efficient clip segmentation, sampling strategies, and multi-threaded pre/post-processing pipelines.
  

Cloud Deployment & Infrastructure

• Deploy models in AWS, GCP, or Azure cloud environments.

• Develop scalable, containerized inference systems (Docker, REST APIs, async queues).

• Build video ingestion pipelines with automated processing and feedback loops.
  

Analytics & Evaluation

• Develop evaluation pipelines for pose estimation (e.g., PCKh, mAP), detection accuracy, and frame-wise precision.

• Integrate scoring models to map visual outputs to business or sports-specific metrics.
  

Requirements

Tech Stack & Tools

• Languages: Python (expert), C++ (optional), Bash scripting

• Frameworks: PyTorch, OpenCV, ONNXRuntime, TensorFlow (ref only)

• Video Tools: FFmpeg, MoviePy, Decord (bonus: NVIDIA DALI)

• Infrastructure: Docker, REST APIs, cloud deployment on AWS/GCP/Azure

• Experiment Tracking: W&B, MLflow, TensorBoard (optional but useful)

Ideal Candidate

• 5+ years of experience in computer vision and deep learning

• Strong track record of delivering production-ready vision pipelines

• Proven ability to work independently and in a cross-functional remote team

• Experience in real-time or near-real-time video inference is a major plus

• Experience in the SportsTech industry or a related field involving human motion analysis.

• Strong track record of delivering production-ready vision pipelines.

• Experience communicating technical concepts to clients and stakeholders.