Vision Analytics AI System Development

via Freelancer ·

Budget / Salary$250–750
TypeFreelance project
LocationRemote
Posted43 minutes ago
I need a production-ready AI system that ingests image or video streams and automatically detects people, objects, activities, quality problems (e.g., damage or safety risks), and returns counts, locations, and movement patterns in real time. The model’s insights will feed directly into my existing dashboards, so a clean REST or gRPC interface is essential.

The core work involves:
• Selecting or designing an architecture that balances speed and accuracy for multi-class, multi-task computer vision (YOLOv8, Detectron2, MMDetection, or a comparable framework are all acceptable).
• Training, fine-tuning, and evaluating the model on a dataset I will supply—mostly 1080p security-camera footage.
• Implementing efficient inference (CUDA/TensorRT preferred) with clear, version-controlled Python code.
• Packaging the solution in Docker with concise documentation and API examples so my team can deploy it on our on-prem server or AWS GPU instances.

Acceptance criteria
• ≥90 % mAP on the validation split for each class group (people/objects, safety risks, counting).
• Inference latency ≤100 ms per 1080p frame on an RTX 4090.
• Endpoint returns JSON with bounding-box coordinates, class labels, confidence, and frame timestamp.
• README that lets a new engineer reproduce training and spin up the container in under 30 minutes.

Fluent, proactive English communication is crucial because we will iterate quickly on edge-case errors and dataset refinements. If you have recent projects showcasing similar vision pipelines or have published benchmark results, share them so I can gauge fit.
java python software architecture cuda c++ programming docker computer vision deep learning rest api
Apply on Freelancer →

Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.