Computer Vision Expert Needed for YOLO Sidewalk Defect Detection and Severity Classification

We are looking for an experienced Computer Vision specialist to build and train two models for our graduation project, “Khota,” within 6 days of starting. Khota helps users report sidewalk defects by uploading photos with their locations. AI identifies the defects and their severity to help authorities prioritize maintenance. Our dataset is already available. Dataset collection is not required. Scope of work: • Build and train a YOLO object detection model to locate and identify Surface Damage and Curb Ramp Issues. • Build and train a separate severity classification model using cropped defect images and our provided classification criteria. • Review dataset readiness and train/validation/test splits to prevent data leakage. • Improve and evaluate both models. We aim for approximately 70% or higher on the appropriate metric for each model, subject to dataset review and feasibility. • Connect the models into a pipeline that accepts an image and returns each defect’s location, type, and severity. Deliverables: • Complete source code for training, evaluation, and inference. • Trained model weights, training settings, and dependencies. • Evaluation report including detection mAP, precision and recall, and classification accuracy, Macro-F1, and confusion matrix. • Setup and usage instructions. • A handover session explaining the implementation and how to run and retrain the models. Required skills: Python, PyTorch, YOLO, OpenCV, object detection, and image classification. We would like brief daily progress updates and clear communication about results or blockers. This scope covers the AI models and inference pipeline, not mobile app or dashboard development. Please include relevant examples of custom model training, your proposed approach, availability, and total cost. Specify whether GPU costs are included and confirm that you can meet the 6-day deadline.