Skip to main content

Quickstart

Start using AI Platform in 5 minutes.
Follow this short guide to go from deploying your first AI model to viewing real-time detections and setting up analytics.

Prerequisites​

  • Cordatus account and access to Web App or Client
  • At least one connected device with Cordatus Client installed
  • At least one active camera added to your device
  • Sufficient GPU resources on the device for inference
  • (For first-time deployment) Stable internet connection for downloading inference engine (35GB)

See Camera Setup and Device Installation for details.

Getting Started in 5 Minutes​

1. Choose a Model or Pipeline​

  • Go to AI Platform → Models from the left menu
  • Browse ready-to-use models (e.g., PeopleNet, TrafficCamNet, FaceNet)
  • Or explore pre-built pipelines like Demographic-Analytics or Plate-Recognition
  • (Optional) Upload your own YOLO model (.pt + labels.txt) or TensorFlow/TAO model
  • (Optional) Create a custom pipeline using the Model Composer drag-and-drop interface
  • Click the ▶ (Play) icon next to the model/pipeline you want to deploy

2. Select Target Device​

  • Choose which device will run the inference job
  • Click Select Device to continue

3. Configure Job Settings​

  • Job Name: Give your job a descriptive name
  • Camera Selection: Select one or more cameras to process
  • GPU Configuration: Choose which GPU to use (if multiple GPUs available)
  • (Optional) Add Alarm: Select pre-defined model-based alarms
  • (Optional) Schedule Job: Set automatic start/stop times
  • Click Start Job to deploy

4. View Real-Time Inference​

  • The stream page opens automatically showing live detection results
  • Detected objects appear with red bounding boxes and labels
  • Toggle detection overlays on/off using the center button
  • View stream metrics (FPS, bitrate, resolution) by clicking the graph icon

5. Add Analytics Rules​

  • Click New Analytic button on the right panel
  • Choose an analytics type:
    • Line Crossing: Draw a line to count objects crossing it
    • ROI Filtering: Draw a polygon to count objects inside/outside
    • Direction Detection: Draw an arrow to detect movement direction
    • Entry & Exit Detection: Draw entry/exit lines to track in/out counts
  • Draw the analytics element on the video frame
  • Give it a label and click Save Without Alarm
  • Watch bounding boxes turn blue when objects trigger the rule
  • (Optional) Add real-time Counters to display analytics results as numbers on screen

6. Set Up an Alarm (Optional)​

  • While adding an analytics rule, click Save With Alarm instead
  • Or create model-based alarms from AI Platform → Alarms → New Inference Alarm
  • Configure:
    • Alarm name and notification channels
    • Alarm rules: frame-based or time-based triggers
    • Target cameras where the alarm should be active
  • Apply alarms when starting a job or add them to running jobs

7. Optimize with ROI (Optional)​

  • On the stream details page, click Model ROIs button
  • Draw one or more regions where you want inference to run
  • Select which models should operate in each ROI
  • This improves performance by focusing on relevant areas only

8. Manage Jobs​

  • Go to AI Platform → Jobs to see all running jobs
  • Click Stop to pause a job (keeps data and settings)
  • Click Restart to resume a stopped job
  • Click Delete to permanently remove a job
  • Jobs can also be managed from Cordatus Client
  • View Child Jobs to see job history and execution records

9. View Analytics Dashboards (Optional)​

  • Go to AI Platform → Dashboard Collections to create custom dashboards
  • Create a collection linked to your model/pipeline
  • Add charts: Distribution (pie), Time-Based (bar/line), Counter, Entry-Exit
  • Customize time ranges and visualize inference analytics

10. Review Alarm Records (Optional)​

  • Go to AI Platform → Alarms → Reports to watch alarm events
  • Select jobs and time range to view triggered alarms
  • Use Timeline to navigate alarm events (blue blocks)
  • Watch recordings with AI inference overlays (bounding boxes and labels)
  • Use Slideshow Mode to quickly review multiple alarms
  • Use Master Synchronization to view the same event from multiple cameras

Expected Result​

  • Model or pipeline successfully deployed on selected device
  • Real-time detection results visible with bounding boxes and labels
  • Analytics rules active and triggering on object behaviors (bounding boxes turn blue)
  • Alarms configured and ready to send notifications
  • Stream metrics showing healthy FPS and performance
  • ROI preprocessing optimizing inference performance
  • (Optional) Custom dashboards displaying analytics data with various chart types
  • (Optional) Alarm records accessible for review with AI overlays and timeline navigation

Next Steps​

  • Explore custom YOLO model upload (v5, v6, v7, v8, v9, v10, v11) with automatic TensorRT optimization
  • Upload TensorFlow or NVIDIA TAO models for custom use cases
  • Create multi-model pipelines using the visual Model Composer (no coding required)
  • Set up multiple alarms with different notification channels
  • Schedule jobs to run automatically at specific times
  • Use analytics cumulative counts for business insights
  • Create custom dashboards with distribution, time-based, counter, and entry-exit charts
  • Review alarm records with AI overlays and slideshow mode
  • Monitor device metrics during inference for optimization
  • Use child jobs to track job execution history

For detailed documentation, see Models, Jobs, Analytics, Alarms, Preprocess, and Watching Alarm Records.