Computer Vision / Edge ML Engineer – Lightweight Medical Image Quality Assessment

Project type: Fixed-price freelance project Duration Approximately 4–6 weeks, part-time Indicative budget $1,800–$2,500 depending on experience and proposed approach Stack Python, OpenCV, NumPy, PyTorch/ appropriate, REST API We are developing a computer-vision system for the analysis of medical images and are looking for a freelance Computer Vision / Edge ML Engineer to develop a lightweight Image Quality Assessment (IQA) module What the module should assess The system should identify quality issues including but not limited to: Image blur / lack of focus, Under-exposure and over-exposure, Excessive glare / reflections,Poor etc. The module should return an interpretable PASS/FAIL decision together with the reason(s) for rejection and, where useful, individual quality scores. Key technical constraint (important) - *Low computational cost is a primary requirement.* The solution should run efficiently on an average CPU and should preferably be suitable for eventual mobile/edge deployment. We are particularly interested in candidates who can determine when classical computer-vision techniques are sufficient and when a small learned model is justified. Solutions dependent on large deep-learning models, GPUs or cloud-based model inference are not appropriate for this project. Potential approaches may include OpenCV-based image statistics, sharpness/frequency measures, histogram and illumination analysis, etc. Expected work and deliverables 1. Review representative labelled medical image data and characterize common quality failures. 2. Benchmark candidate approaches for each quality criterion. 3. Develop the image-quality assessment pipeline. 4. Determine and validate appropriate thresholds/classification criteria. 5. Optimize the implementation for CPU/edge execution. 6. Evaluate performance on a held-out dataset. 7. Benchmark inference latency, memory consumption and model size. 8. Deliver clean, maintainable Python code with appropriate tests andtechnical documentation. 9. Package the completed IQA module as a deployable API/microservice that can be integrated into an existing application architecture. The API should: Accept an image as input. Run the image-quality assessment. Return a structured response containing PASS/FAIL status, identified quality problems and relevant quality scores. Include appropriate error handling and input validation. Be documented sufficiently for integration by another developer. Be containerized (preferably Docker) so that it can be deployed as an independent microservice. Run without requiring GPU infrastructure. Desired experience * Strong Python and OpenCV experience * Practical computer vision/image-processing experience * Understanding of image sharpness, exposure, illumination and image-quality metrics * Experience building computationally efficient CV pipelines * PyTorch experience * Experience building REST APIs in Python (FastAPI preferred) * Docker/containerization experience * ONNX, TensorFlow Lite or mobile/edge ML experience is highly desirable Medical-imaging experience is useful but **not required**. ### Definition of successful delivery At completion, we expect to receive: * A validated image-quality assessment pipeline * Performance results on a held-out test dataset * Per-quality-criterion evaluation metrics * CPU inference-time and resource benchmarks * Source code and configuration * Trained model weights, if learned components are used * Tests and technical documentation * A documented, containerized REST API that can be deployed as an independent microservice * Instructions for running the service locally and deploying it into an existing application environment All source code, trained models and project-specific implementation produced under the engagement must be delivered as part of the project. ### When applying Please provide: 1. 1–3 examples of relevant computer-vision projects you personally implemented. 2. Your experience with OpenCV and lightweight/edge ML. 3. Your experience deploying ML/CV models as APIs or microservices. 4. A brief description of how you would approach blur, exposure and incorrect-view detection while minimizing computational requirements. 5. Your proposed timeline. 6. Your fixed-price quotation. Shortlisted candidates may initially be offered a **small paid technical milestone/proof-of-concept** before proceeding with the complete project. Please do not submit generic AI/LLM portfolios. We are specifically looking for hands-on computer vision, image processing and lightweight ML engineering experience.