Configure P4 Detect

You can configure P4 Detect to use different object labels, label groups, or batch size.

After you have configured P4 Detect, you can connect it to P4 Search. See Set up auto-tagging with P4 Detect.

Update the P4 Detect configuration files

  1. Update the configuration files in the /label_config directory.

  2. Restart P4 Detect:

    docker stop p4detect.helix

    docker start p4detect.helix

Sample configurations

The P4 Detect Docker container comes with sample configuration files in the /label_config directory.

general.json contains a list of image labels:

{
    "_comment": "general",
    "labels": [
        "airplane",
        "alien",
        "ally",
        "ammo",
        "animal",
        "archer",
        "assassin",
        "axe",
        "barrel",
        "beach",
        "bear",
        ...
    ]
}

You can have multiple image label configuration files.

label_groups.json contains the image label configuration files to include when running P4 Detect:

{
    "general": "general.json"
}

batch_sizes.json contains the reading batch size for each image label configuration file:

{
    "general": 32
}

Consider using a small batch size (such as 32) when the number of image labels is very large, or if you have low GPU capacity.

Next steps