Skip to content

Launch CVAT Job

Prerequisites#

Before you start the launch process, you need to meet these prerequisites:

  • get some POL for your Job Launcher wallet (using Amoy testnet faucets); you will need it to create escrows while launching jobs

Info

In Amoy, it's recommended to have at least 1 POL to satisfy any need of the Job Launcher.

  • get some HMT for your wallet (you can use our Faucet app); you will need it to fund jobs you launch
  • go to the Job Launcher UI and create an account as the Job Requester; top up your account with HMT (100 should be enough to run multiple jobs)

Info

If you don't have an option to choose "Crypto" when launching jobs or to top up your account, add your user to the whitelist in the corresponding Job Launcher DB table.

Launch sample CVAT job#

The repository has sample dataset images and ground truth files that let you run an image_boxes (aka "Bounding Boxes") job in the packages/examples/cvat/exchange-oracle/tests/utils/datasets folder. To launch the job, upload the necessary files to S3-compatible storage.

For a local setup, you can use Minio running on your machine:

  1. Open the Minio UI and log in as super admin.
  2. Create a local-dataset bucket and make it Public.
  3. Upload instances_default.json to the root of the bucket.
  4. Create an images-folder in the bucket; upload images from the datasets folder to it.

Info

You can use your public S3 bucket for that purpose, then you will be able to use Job Launcher UI to run jobs.

Then open the Job Launcher server Swagger docs and follow these steps:

  1. Use /auth/signin to get a JWT access token for your Job Requester account.
  2. Authorize Swagger with that access token.
  3. Go to the /job/cvat endpoint and run it with the following body:
{
  "chain_id": 80002,
  "qualifications": [],
  "payment_currency": "hmt",
  "payment_amount": 12,
  "escrow_fund_token": "hmt",
  "requester_description": "Local CVAT job using Minio to host dataset",
  "data": {
    "dataset": {
      "provider": "local",
      "bucket_name": "local-dataset",
      "path": "images-folder"
    }
  },
  "labels": [{
      "name": "cat"
  }],
  "min_quality": 0.2,
  "ground_truth": {
    "provider": "local",
    "bucket_name": "local-dataset",
    "path": "instances_default.json"
  },
  "user_guide": "https://humanprotocol.org",
  "type": "image_boxes"
}

Info

If necessary, adjust values in the request body (e.g., if you have different paths).

Then go to the Job Launcher UI, where you should be able to track the job status.