Following a significant update in december 2024 to the Deep Live Hub of our DeepVA composite AI platform last year, we are pleased to kick off the year with two brand new AI services for the Deep Media Analyzer.
New service: Visual Understanding
The Visual Understanding Module is a recent addition to Aiconix’s Deep Media Analyzer, designed to enhance the analysis of images and videos through prompt-based visual language understanding. Designed for users who require a deeper understanding of their visual content, this module complements our already powerful zero-shot object and scene recognition with its customization.
Key Features
Visual Language Comprehension: Interprets visual elements such as symbols, images, and layouts to provide insights without relying on audio information.
Prompt-Based Analysis: Users input specific prompts to guide the analysis, enabling tasks like scene description, content summarization, emotion and tone analysis, highlights extraction and audience engagement insights.
Since it is based on a prompt, there are many different use cases possible, here are some example prompts:
The results are returned in as text together with the initial prompt. This Module is available for all DeepVA Users and we will be adding further features and parameters to this module in the next releases.
New feature: Threat Detection
Designed primarily for security applications such as monitoring of large sites or buildings, this module can be integrated into workflows via an API for automated threat detection and response. It is a visual language understanding model, pre-trained to recognize dangerous situations such as violence, medical issues or hazards such as fire, to provide faster alerts to CCTV operators.
The results are returned in a clear yes or no indicator, along with a written explanation of why the system reached that judgement. This allows a human operator to check the video resource more quickly and take action or call for further assistance if required. This module is available to all DeepVA users and is ideally used via the API.
Minor Improvements: Increased Usability
- We change the default model for our Object and Scene Recognition in the Visual Mining Wizard to the Zero-Shot model, which is our state-of-the-art model and can even be customised using dictionaries.
- We refactored the Help Center page in our API in order to provide faster and better guidance, including the linking of our new Knowledge Hub and Support Form.
- Improved UI of speaker dataset label.
Updates for the Deep Live Hub
In December 2024 Changelog, we already introduced significant updates to the Deep Live Hub of our composite AI platform, DeepVA, enhancing usability, flexibility, and security.
Highlights include the integration of multi-streaming, enabling simultaneous live streaming to multiple platforms such as YouTube Live, Facebook Live, and Twitch without requiring external tools like Restream IO. Other additions include reusable subtitle settings, enhanced security through RTMPS, and improvements in stability and error handling.
This week we further improved the Deep Live Hub by adding:
SRT (Secure Reliable Transport) Push & Input
Further enhanced security through RTMPS and SRT (Secure Reliable Transport), for more details see our last changelog.
Loadbalacing
We also improved the load balancing to deliver good results with a lower latency Switching the load balancing for lower latency.
Roadmap
Next month we will be releasing new features including deeper text understanding and more functions for the Visual Understanding module, all tied to the ability to combine the modules of our Composite AI platform to achieve higher levels of automation. Stay tuned!
All DeepVA changelog updates are available here: https://docs.deepva.com/changelog/
