Healthcare AI MLOps

Healthcare Image Processing

An end-to-end pipeline that turns Healthcare video into deduplicated frames, object detections, and standard annotations—ready for downstream ML—with production-minded metrics and container packaging.

View on GitHub Back to Healthcare projects
Diagram: video through dedupe, YOLO, COCO, metrics

Overview

The pipeline extracts frames from video, removes near-duplicates with perceptual hashing, runs YOLOv8 detection, emits COCO-format annotations, and publishes operational metrics. It is designed for repeatable runs, clear reporting, and integration with common monitoring stacks.

Capabilities

Frame intelligence

Configurable perceptual-hash deduplication cuts redundant storage and compute while preserving clinically distinct frames.

Detection & labels

YOLOv8 inference with tunable confidence; outputs map cleanly into COCO-style annotations.json for training and evaluation tooling.

Observability

Prometheus counters and histograms for throughput, detections, and timing—suitable for dashboards and alerting.

Operational packaging

Dockerfile, Makefile targets, YAML configuration, and a markdown pipeline report summarising stats, timing, and class distribution.

Visual summary

Diagram: Prometheus metrics, COCO exports, and pipeline reporting
Reporting and monitoring: Prometheus metrics alongside COCO exports and generated reports. The high-level pipeline (video through deduplication, detection, and metrics) is shown in the hero above.

Technology stack

  • Python 3.9+
  • YOLOv8
  • OpenCV / FFmpeg
  • Perceptual hashing
  • Prometheus
  • Docker
  • pytest

Need a similar vision or video ML pipeline?

Get in touch