Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hydra: Computer Vision for Data Quality Monitoring

Thomas Britton, Torri Jeske, David Lawrence, Kishansingh Rajput

TL;DR

Hydra addresses the burden of manual data quality monitoring by delivering near real-time, computer vision–based QC across Jefferson Lab halls. It combines a Python back-end (with TensorFlow models and a MySQL database) and a web front-end to manage, visualize, and interact with model inferences, labeling, and historical data. Key contributions include a modular inference pipeline with ZeroMQ-based messaging, gradCAM explanations for Bad classifications, hall-agnostic deployment, and an integrated labeling and evaluation workflow. The system enables faster issue detection and decision-making, with ongoing developments aimed at expanding detection capabilities, improving human control, and enhancing computational efficiency.

Abstract

Hydra is a system which utilizes computer vision to perform near real time data quality management, initially developed for Hall-D in 2019. Since then, it has been deployed across all experimental halls at Jefferson Lab, with the CLAS12 collaboration in Hall-B being the first outside of GlueX to fully utilize Hydra. The system comprises back end processes that manage the models, their inferences, and the data flow. The front-end components, accessible via web pages, allow detector experts and shift crews to view and interact with the system. This talk will give an overview of the Hydra system as well as highlight significant developments in Hydra's feature set, acute challenges with operating Hydra in all halls, and lessons learned along the way.

Hydra: Computer Vision for Data Quality Monitoring

TL;DR

Hydra addresses the burden of manual data quality monitoring by delivering near real-time, computer vision–based QC across Jefferson Lab halls. It combines a Python back-end (with TensorFlow models and a MySQL database) and a web front-end to manage, visualize, and interact with model inferences, labeling, and historical data. Key contributions include a modular inference pipeline with ZeroMQ-based messaging, gradCAM explanations for Bad classifications, hall-agnostic deployment, and an integrated labeling and evaluation workflow. The system enables faster issue detection and decision-making, with ongoing developments aimed at expanding detection capabilities, improving human control, and enhancing computational efficiency.

Abstract

Hydra is a system which utilizes computer vision to perform near real time data quality management, initially developed for Hall-D in 2019. Since then, it has been deployed across all experimental halls at Jefferson Lab, with the CLAS12 collaboration in Hall-B being the first outside of GlueX to fully utilize Hydra. The system comprises back end processes that manage the models, their inferences, and the data flow. The front-end components, accessible via web pages, allow detector experts and shift crews to view and interact with the system. This talk will give an overview of the Hydra system as well as highlight significant developments in Hydra's feature set, acute challenges with operating Hydra in all halls, and lessons learned along the way.
Paper Structure (18 sections, 5 figures)

This paper contains 18 sections, 5 figures.

Table of Contents

  1. Introduction
  2. The Hydra Back-end
  3. Database
  4. Feeder
  5. Load Balancing
  6. Predict
  7. Keeper
  8. The Hydra Front-End
  9. Labeler
  10. Model Training Report Page
  11. Library
  12. Status
  13. Run
  14. Log
  15. Grafana
  16. ...and 3 more sections

Figures (5)

  • Figure 1: A simplified diagram of the Hydra system. The orange arrows indicate the path of inference orders through Hydra's back end processes. The MySQL database is indicated in green.
  • Figure 2: Snapshot of the online Labeler. Users can select the plot to label, select the appropriate label and label thousands of images quickly.
  • Figure 3: The library page displaying the enhanced confusion matrix for the active model (indicated with the star icon) is shown.
  • Figure 4: The Hydra Run page displaying near real time classifications along with gradCAM heatmaps overlayed on the images. These help indicate problematic areas of the images to aid the shift crew in diagnosis.
  • Figure 5: The Hydra Log page displaying confirmed bad plots from the previous day.