Using Custom Taxonomies

For clients with specific tagging requirements, adopting a custom taxonomy can be transformative for the management and classification of asset collections. Our Custom Classification service uses taxonomy-based training to finely tune our tagging models to the unique context of a collection, significantly enhancing tagging relevancy.

Case Study: Video Game Environments

Consider a challenge faced by a client in the gaming industry. They have an extensive collection of assets for various environmental elements like terrain, flora, weather conditions, and times of day.

To streamline their tagging process, we create a custom Environmental Element Taxonomy. This taxonomy efficiently classifies assets based on specific environmental characteristics.

Taxonomy Structure

Terrain: Mountains, Forests, Plains, Urban, Water Bodies.
Flora: Trees (Pine, Oak, Birch), Bushes, Grass Types, Flower Varieties.
Weather Conditions: Sunny, Rainy, Snowy, Foggy, Stormy.
Time of Day: Dawn, Morning, Noon, Evening, Night.

As structured data, this would look like:

{
  "EnvironmentalElements": {
    "Terrain": ["Mountains", "Forests", "Plains", "Urban", "Water Bodies"],
    "Flora": {
      "Trees": ["Pine", "Oak", "Birch"],
      "Bushes": [],
      "GrassTypes": [],
      "Flowers": []
    },
    "WeatherConditions": ["Sunny", "Rainy", "Snowy", "Foggy", "Stormy"],
    "TimeOfDay": ["Dawn", "Morning", "Noon", "Evening", "Night"]
  }
}

With the introduction of this taxonomy, an asset processed by our model will receive a tag from each relevant class. A 3D model of a pine tree located in a forest setting on a rainy evening would be labeled as follows:

Terrain: Forests  
Flora: Trees: Pine  
Weather Conditions: Rainy  
Time of Day: Evening

This approach allows for efficient sorting and searching of environmental assets, enabling the precise selection of appropriate assets for scenes where attention to detail is paramount.

Implementation

🏷️

Do you have tagging needs that require the use of a custom taxonomy? Book a call, or send us a message.

The implementation of classification using custom taxonomies follows these broad steps:

Consultation: Initial consultation with our in-house ML team to define a taxonomy structure that aligns with your project’s vision. Already have a taxonomy? We will gladly integrate it.

Custom Training: We will rigorously fine-tune our tagging model to your custom taxonomy, meeting agreed upon accuracy and precision specifications.

Integration: We support you with integrating the trained model into your asset management system. You can find out more about the various deployment processes we support here.