Hodavh/AnimalDetect-African-Wildlife-Detection-Platform

A production-ready MLOps pipeline for real-time African wildlife object detection using Flask, TensorFlow Serving, and Docker, with GPU-accelerated inference via gRPC and a full MLOps metrics dashboard. # AnimalDetect — African Wildlife Detection Platform > A production-ready MLOps pipeline for real-time African wildlife object detection using Flask, TensorFlow Serving, Docker, Prometheus, and Grafana, with GPU-accelerated inference via gRPC and a full MLOps observability stack. **Tech Stack**: TensorFlow Serving • Flask • Docker • Docker Compose • Faster R-CNN ResNet50 • gRPC • Prometheus • Grafana • Python • PIL --- ## Problem Statement Deploying machine learning models in enterprise environments requires more than just a trained model — it demands a robust, scalable, and observable serving infrastructure. This platform demonstrates a production-grade MLOps pipeline where a custom-trained Faster R-CNN ResNet50 model for African wildlife detection is served via TensorFlow Serving in a GPU-accelerated Docker container, consumed by a Flask web application through gRPC, and monitored via a real-time MLOps observability stack built with Prometheus and Grafana. The system is fully containerised using Docker Compose, enabling reproducible deployments across any environment. --- ## Key Features - **Custom wildlife model**: Faster R-CNN ResNet50 trained on three African wildlife species - **GPU-accelerated inference**: TensorFlow Serving runs on NVIDIA GPU via Docker with full CUDA support - **gRPC communication**: High-performance binary protocol between Flask and TensorFlow Serving (port 8500) - **Prometheus monitoring**: Automatically scrapes inference metrics from Flask every 15 seconds - **Grafana dashboards**: Real-time visualisation of inference time, detection counts, and confidence scores - **Side-by-side visualisation**: Original vs. classified image comparison with PIL-rendered bounding boxes - **File validation**: Upload restricted to .jpg, .jpeg, .png with user-friendly error messages - **Containerised microservices**: Four-container architecture orchestrated via Docker Compose - **Persistent image storage**: Original and classified images saved to se …