Clarifai will review your list of concepts, propose taxonomy structure options, and suggest concept changes, including separating or combining hierarchical concepts.
There are a number of ways to configure a model to get the best results. Clarifai’s team will present several options and explain the pros and cons of each possibility. A few examples of the type of model options we might suggest include:
- Creating sub-models to improve the accuracy of several concepts that are very visually similar
Deciding whether to implement the concepts mutually exclusive setting (whether you want more than one concept to be returned per image)
- Creating a “closed environment” - In layman's terms, this means that if there's a chance that your model will see uploads from images that aren't necessarily in your training set, generally this should be set to false. If you expect to run a trained model on images that do not contain any of the concepts in your model, then we would set closed_environment to false (the default).
- Optimizing custom hyperparameters to achieve maximum model performance.
Data Cleansing & Curation
This step prepares the data for use in training our models. Clarifai’s Data Strategists can discard improper data, recognize important variables or visual features, and check for data imbalances that could impact the predictions of the model.
Model Training Setup
Clarifai can create different models and different versions of each model in the cloud and deploy to your instance, or we can configure your environment to train on-premise.
Model Performance Evaluation
Clarifai can create different models and different versions of each model and run evaluation procedures and present you with performance metrics of each of the model versions.