Skip to main content

Filter Demo

This guide explains how to set up and run the EVP License Plate Filter Demo using Docker, Google Cloud, and optional GPU acceleration.

Prerequisites​

Google Cloud Setup​

All Docker images and assets are hosted on Google Artifact Registry (GAR).

  1. Obtain your GOOGLE_SERVICE_ACCOUNT_CREDENTIALS.json credentials file from Plainsight.

  2. Authenticate:

    gcloud auth activate-service-account --key-file=GOOGLE_SERVICE_ACCOUNT_CREDENTIALS.json

  3. Configure your environment:

    gcloud config set project plainsightai-prod
    gcloud auth configure-docker us-west1-docker.pkg.dev

Downloading Demo Assets​

Retrieve the full demo bundle:

gcloud artifacts generic download \
  --project=plainsightai-prod \
  --location=us-west1 \
  --repository=files \
  --package=evp-demo-getting-started \
  --version=v1.1.2 \
  --destination=.

Unzip the assets:

unzip evp-demo-getting-started.zip

You should see:

video.mkv
model.zip
docker-compose.yaml
docker-compose-local.yaml
docker-compose-gpu.yaml
docker-compose-local-gpu.yaml
.env
README.md
crop_filter/
filter_license_plate_detection/
filter_optical_character_recognition/
filter_license_annotation_demo/

Running the Demo​

Standard CPU Run​

docker compose -f docker-compose.yaml up

This launches:

  • vidin → streams video
  • filter_license_plate_detection → detects license plates
  • crop_filter → crops plates
  • filter_optical_character_recognition → Runs OCR (Optical character recognition) on plates
  • filter_license_annotation_demo → overlays results
  • webvis → web visualization (localhost:8002)

Running with GPU Acceleration​

If you have a GPU and proper Docker setup:

docker compose -f docker-compose-gpu.yaml up

This enables GPU usage for inference filters (filter_license_plate_detection and filter_optical_character_recognition).

Development Mode (Editable Filters)​

For live-editing Python code (filter.py) without rebuilding images:

docker compose -f docker-compose-local.yaml up

Each service mounts the local filter source directly.
Restart an individual filter to apply changes:

docker compose restart filter_license_annotation_demo

Development + GPU Mode​

For live-editing and GPU acceleration:

docker compose -f docker-compose-local-gpu.yaml up

Folder Structure​

video.mkv                        # Input video
model.zip                        # Detection model
docker-compose.yaml              # Standard demo run
docker-compose-local.yaml        # Editable mode
docker-compose-gpu.yaml          # GPU run
docker-compose-local-gpu.yaml    # Editable + GPU run
crop_filter/filter.py
filter_license_plate_detection/filter.py
filter_optical_character_recognition/filter.py
filter_license_annotation_demo/filter.py
README.md
.env

Tips​

  • Access Web UI at http://localhost:8002
  • View live logs: 
    docker compose logs -f filter_license_annotation_demo
  • Restart a single filter: 
    docker compose restart filter_license_annotation_demo

Troubleshooting​

ProblemSolution
Blank or frozen videoEnsure all filters are running
Missing OCR textCheck topic connection (cropped_main)
Docker pull failuresConfirm GCP credentials and artifact access
GPU not detectedVerify Docker + NVIDIA runtime setup

License Plate Annotation Behavior​

  • Plates are detected and cropped
  • OCR is performed on cropped plates
  • Cleaned OCR results (e.g., ABC1234) are overlaid
  • Last seen license plate is used if OCR temporarily fails
  • Cropped plate image is inserted into the main video

Summary​

This demo showcases a complete event-driven vision pipeline — real-time license plate detection, OCR, annotation, and live web visualization — all powered by Plainsight's modular filter architecture.

You can extend it with your own videos, models, or downstream analytics!