Skip to content
English
Create Support Ticket
  • There are no suggestions because the search field is empty.

What is the Deep Model Customizer?

Easily Create and Customize AI Models with Minimal Data and Effort

The Deep Model Customizer is a powerful no-code AI tool designed to help users create their own customized AI models with minimal effort. Unlike pre-trained models that may not recognize unique or specific objects in your content, the Deep Model Customizer allows you to train AI models that suit your exact needs. This tool drastically reduces the time, cost, and complexity typically involved in AI training.

Note:
The Deep Model Customizer is currently focused on file-based analysis, but in the coming months modules will also be available for live resources, which will be integrated into the Deep Live Hub.

The Deep Model Customizer incorporates several modules:

  • Custom Face Training:

    • Easily train the system to recognize new faces with just a few images. This is particularly useful for building custom facial recognition datasets to suit your needs.

  • Landmark Training:

    • Quickly train the AI to recognize specific landmarks or architectural structures, making it ideal for geographic content or identifying famous buildings in your footage.

  • Logo Training:

    • Train the system to recognize company logos in videos or images, which can help you monitor brand visibility and track intellectual property usage in media.

  • Speaker Training:

    • You can train the AI to recognize individual speakers from your footage, even with minimal data. This is particularly useful for media and news organizations that need to identify public figures or panelists.

Few-Shot Learning:

Few-Shot Learning enables the Deep Model Customizer to train accurate AI models with just a small amount of data, reducing the typical time and data requirements for AI training.

 

Other training methods for building custom models:
With the Deep Model Customizer, DeepVA offers additional methods to create custom datasets more efficiently, some of which include partially automated processes. These methods significantly reduce the time and effort required to collect and prepare training data for your AI models.

  •  Automate data collection with Deep Collector:
    The Deep Collector automates the collection of training data from pre-tagged video footage. For example, if your footage contains lower thirds (on-screen text), the system can automatically identify and extract relevant data for training purposes. This automated process reduces the time needed to create a dataset by 90%, making it much faster to build a comprehensive, labelled dataset.
  • Indexing with Deep Indexer:
    The Deep Indexer provides a more structured way of organising and managing large amounts of media data. It works by indexing all the voices and faces in your media assets and assigning them unique IDs. These IDs can then be manually renamed, allowing the system to recognise and track these individuals in future analysis.

    The Deep Indexer enables additional informations for various assets: 
    • Past assets: All older assets analysed by the Deep Indexer are indexed and manually renamed IDs are given the appropriate label.
    • New assets: The system will automatically apply the correct labelled names to previously indexed individuals.

Use Cases:

  • Media Companies: Use it to create face or speaker recognition datasets for identifying people in footage.
  • Brand Monitoring: Train AI models to recognize logos, ensuring that your brand's presence is consistently tracked across all media.
  • Cultural Institutions: Museums or historical archives can use the tool to recognize specific landmarks or artifacts in their footage.

Privacy & No Vendor Login:

The Deep Model Customizer gives you full control over your AI models and data. This is only accessible to members of your account, no other user, third party applications or we have access to your models or training data. The data belongs to you and remains yours.
However, you can export your data and models at any time to avoid platform dependency,
it is even possible to manually share generated models across various users or license them
to other clients, if requested.
Click here for more details on privacy and data security.