In Developers, Computer Vision, Artificial Intelligence, Deep Learning, Models

Introducing Clarifai Deep Training

By Jeff Toffoli

H1

H2

Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time. Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.

video-scene-detection-car

Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time. Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.Seamlessly integrate our software in any environment, including smart devices, environments with low-computational power and limited or no internet access with zero lag-time.

Since 2013 Clarifai has offered dozens of pre-trained models that make it easy to get started in AI. Plus, we offer award winning technology for custom training, which lets people build their own custom models on top of pre-trained Clarifai base models. Our custom training technology helps companies of all sizes build high performing AI models quickly, with less training data. 

Today, we are announcing Deep Training. With Deep Training data scientists and engineers can build their own "base models" that work just like Clarifai Models in our platform.

What is Deep Training? 

Deep training is a process that builds a custom neural network for your application from the ground-up. This means that your model can become an expert in recognizing the unique set of visual features that is important in your data set.

When is Deep Training better than Custom Training? 

There are a couple cases when deep training is preferable to custom training on top of Clarifai Models.

1. When your data is unique 

There are times when your dataset and use case are unique and require your model to recognize a different set of visual features than the ones recognized by pre-trained models. This is common in scenarios when you're working with visual data of a highly technical or specialized nature, or when the distribution of data the base model was trained on is materially different than the data you expect to infer on.

uncommon_feature

 

2. When you need to maximize model accuracy

For many users—especially those who use AI models to drive some form of automation—model accuracy can be critical. Deep trained models can help you push the accuracy of your models to the limits. Even images of everyday events and objects can vary a lot when you really need to pay attention to the details.  

retail-planogram-retail-store-min

 

How is Deep Training different than using pre-trained Clarifai models?

We've built proprietary technology that allow our users to transfer the feature-recognizing power of Clarifai Models to new custom models. This technique has proven incredibly effective across a wide variety of use cases, and is the starting point for most users on the Clarifai platform. 

Clarifai Models are designed to be general purpose feature identifiers that give you a head start with the most common visual classification and detection tasks. They work best on "everyday" photos and videos, taken with similar cameras, at a similar scale, with a similar set of subjects. With Deep Training, you can custom tailor the feature recognition characteristics of your model to your unique data set and use case.

Who should use Deep Training?

Consider deep training if you have:

  • A custom tailored dataset with at least 1,000 inputs
  • Accurate labels
  • Expertise and time to fine tune model hyperparameters
  • A need to build many custom models leveraging deep trained models

microscopy

Deep Training is in early preview with some of our customers. If you want to start using it or learn more, please contact us

 

Subscribe to updates

Recent Posts