Computer vision custom yolo object detection and web app integration

We are looking for an experienced Computer Vision & AI Software Engineer to develop, fine-tune, and deploy a custom real-time object detection system using YOLOv8. The goal of this project is to build an end-to-end solution—from fine-tuning a YOLOv8 model on custom datasets to deploying a lightweight backend API and Web interface for live inference (image/video/webcam streams). Key Responsibilities • Data Preparation & Fine-Tuning: Preprocess, augment, and split custom image/video datasets to fine-tune a YOLOv8 model (e.g., target classes such as PPE compliance: Hardhat, No-Hardhat, Safety Vest, No-Safety Vest, Mask, Person, etc.). • Model Training & Evaluation: Train and evaluate the model using PyTorch and Ultralytics YOLOv8. Monitor loss curves and report metrics including [email protected], [email protected]:0.95, Precision, and Recall. • Backend API Development: Build a clean FastAPI backend endpoint (/predict) to handle image upload / video frame stream requests and return inference results with bounding boxes (e.g., encoded as Base64 or JSON bounding box formats). • Frontend Web & Live Stream Integration: Create a simple HTML/JavaScript frontend supporting image uploads, local video file processing, and live webcam stream detection. • Model Optimization & Export: Export the trained model to production-ready formats (e.g., ONNX, TensorRT, or .pt) and document deployment steps. Required Skills & Qualifications • Machine Learning / CV: 2+ years experience in Object Detection, Deep Learning, and Computer Vision (OpenCV, PyTorch, Ultralytics YOLOv8). • Backend Development: Strong proficiency in Python and FastAPI. • Frontend Basics: Basic understanding of JavaScript/HTML for handling web assets and canvas bounding box rendering. • Model Evaluation: Ability to deliver detailed evaluation reports (Confusion Matrix, PR curves, F1 curves). Deliverables 1. Trained Weights & Model Checkpoints: Optimized .pt and ONNX weights hosted or uploaded to Hugging Face / GitHub releases. 2. Training Artifacts & Notebook: Annotated Jupyter notebook (YOLO_Training.ipynb) showing dataset prep, loss curves, confusion matrix, and evaluation metrics. 3. End-to-End Application Codebase: o Backend (backend/app.py): FastAPI inference server. o Frontend (index.html, webcam.html): Web interface for upload and real-time webcam monitoring. o Local Script (local/webcam_detection.py): Python script for local live webcam stream testing.