Vertex AI Test

Processes a Planet satellite imagery zip file: unzips, validates integrity, slices GeoTIFFs into PNG tiles, and uploads them to Google Cloud Storage. (Vertex AI batch inference coming soon.)

Build

go build ./...

Requires the tiff2rgba and tiffcrop CLI tools (from libtiff) at runtime. If using Nix, nix develop provides them.

GCP prerequisites

The service account used for authentication needs the following IAM roles:

Role	Purpose
roles/storage.objectAdmin	Read/write objects in the GCS bucket
roles/aiplatform.user	Create and manage Vertex AI batch prediction jobs

Grant them via:

gcloud projects add-iam-policy-binding <PROJECT> \
  --member="serviceAccount:<CLIENT_EMAIL>" \
  --role="roles/storage.objectAdmin"

gcloud projects add-iam-policy-binding <PROJECT> \
  --member="serviceAccount:<CLIENT_EMAIL>" \
  --role="roles/aiplatform.user"

The <CLIENT_EMAIL> is the client_email field from your service account JSON key file.

Usage

vertex-ai-test [flags] <input.zip>

Required flags

Flag	Env var	Description
-bucket	GCS_BUCKET	GCS bucket name for tile uploads
-project	VERTEX_PROJECT or GOOGLE_CLOUD_PROJECT	GCP project ID
-model	VERTEX_MODEL	Vertex AI model ID (numeric ID from the console, not the display name)

Authentication

Supply GCP credentials via one of:

• -creds flag: path to a GCP service account JSON key file
• GOOGLE_APPLICATION_CREDENTIALS env var: path to the same JSON key file
• ADC fallback: if neither is set, the GCE/GKE metadata server is tried

Example using a service account key file:

./vertex-ai-test \
  -bucket my-bucket \
  -project my-gcp-project \
  -model 1234567890123456789 \
  -creds /path/to/service-account.json \
  scene.zip

Or equivalently with env vars:

export GCS_BUCKET=my-bucket
export VERTEX_PROJECT=my-gcp-project
export VERTEX_MODEL=1234567890123456789
export GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
./vertex-ai-test scene.zip

Optional flags

Flag	Default	Description
-tilearea	1000.0	Target tile area in square meters
-gcs-prefix	""	Optional prefix within the GCS bucket
-location	us-central1	GCP region for Vertex AI (or VERTEX_LOCATION env)
-machine-type	n1-standard-4	Batch prediction worker machine type
-max-workers	10	Max prediction worker count
-batch-size	64	Batch size per worker
-debug	false	Enable debug-level logging
-force	false	Re-upload tiles that already exist in GCS (by default they are skipped)
-j	8	Number of parallel GCS uploads

Output

• Tiles are extracted to ./tmp/<scene-name>/tiles/<itemID>/<assetType>/<N>px/tile_X_Y.png
• Uploaded to GCS at gs://<bucket>/[<prefix>/]<catalogID>/<itemID>/<assetType>/<N>px/tile_X_Y.png