Vision-Language Model Training
We're excited to offer two half-day in-person vision-language model (VLM)
trainings at the 2026 Embedded Vision Summit!
Intro Course
Vision-Language Models for Computer Vision Applications: A Hands-On Introduction
Wednesday, May 13, 2026, 9 am - noon
Advanced Course
Vision-Language Models for Video Understanding and Agentic AI
Wednesday, May 13, 2026, 1:30 - 5:00 pm
Presented by:
See below for course details (click either Intro or Advanced tab).
Vision-Language Models for Computer Vision Applications: A Hands-On Introduction
(Looking for the Advanced course? Click the tab “Advanced Course Description” above!)
Are you an engineer, developer or engineering manager eager to harness the power of generative AI for cutting-edge computer vision applications? Join us for an intensive three-hour training session designed to introduce the latest techniques in vision-language models (VLMs) and their integration with traditional computer vision methods. With a focus on the practical application of these techniques for real-world use cases, this course is tailored for professionals looking to expand their skill set in AI-driven computer vision, particularly in systems designed for deployment at the edge.
What You’ll Learn
Introduction to VLMs and LLM+Computer Vision Techniques with Phil Lapsley, Vice President, Edge AI and Vision Alliance:
- We’ll start with an overview of vision-language models and how they differ from conventional convolutional neural networks. We’ll then discuss the advantages and potential drawbacks of integrating LLMs and VLMs with computer vision and explore real-world applications that benefit from these advanced techniques.
Technical Deep Dive with Dr. Satya Mallick, CEO of OpenCV:
- Gain insights into the basics of VLMs, including embeddings, CLIP and how different modalities (text, vision) are encoded. Learn about the types of training data required and the loss functions used in these models. This segment will provide the necessary background to tackle the practical examples that follow.
- First Hands-On Example: Zero-Shot Image Classification. Our first practical example will be image classification with CLIP for zero-shot learning. You’ll build an image classifier capable of recognizing a wide array of images without prior training. Discover how CLIP’s zero-shot classification can be deployed on mobile devices and learn how to fine-tune the model for enhanced performance on specific datasets.
- Second Hands-On Example: VLM with Agnostic Object Detector. We’ll develop a VLM-based visual AI system that identifies objects and reasons about them using pre-existing world knowledge. We’ll accomplish this by using a CNN-based class-agnostic object detector and integrating it with a VLM to answer complex questions about detected objects.
Who Should Attend
This training is ideal for engineers, developers, engineering managers and CTOs with a basic understanding of Python, Jupyter Notebook and computer vision concepts. Whether you’re working in mobile development, embedded systems or cloud applications, this course will provide you with the tools and knowledge to implement sophisticated AI solutions in your projects.
To make the most out of this training, you should have:
- Working knowledge of Python
- Basic familiarity with Jupyter Notebook or Google Colab
- Basic familiarity with computer vision; familiarity with OpenCV and PyTorch is helpful but not required
Why Attend?
The field of generative AI and multimodal LLMs is moving at a truly breakneck pace. This course offers a great way to keep up with this rapidly evolving technical landscape. In particular, it provides a unique blend of foundational knowledge and practical applications, ensuring you leave with actionable skills and access to sample code for continued learning.
Register Today
Registration is $495. Don’t miss this opportunity to enhance your skills and stay at the forefront of computer vision technology. Register today to secure your spot in this transformative training session. (You can save $50 if you also register for the Embedded Vision Summit!)
Vision-Language Models for Video Understanding and Agentic AI
(Looking for the Intro course? Click the tab “Intro Course Description” above!)
Ready to move beyond CLIP-style “image + caption” alignment and build vision-language systems that reason over time and take actions? This live, hands-on advanced training dives into modern VLMs designed for video understanding, multimodal reasoning, and agentic behavior. You’ll learn how today’s “thinking” vision models differ from classic VLM pipelines—and you’ll implement practical patterns for building applications that observe, reason, decide, and adapt.
This session is taught by Satya Mallick, CEO of OpenCV, and focuses exclusively on advanced, real-world capabilities using state-of-the-art models such as Qwen3.5, GLM-4.1V-Thinking, and LLaVA-NeXT.
This is an advanced course. If you haven’t taken the introductory VLM session, we strongly recommend it—or ensure you meet the prerequisites below.
What You’ll Learn
Advanced VLMs: From Perception to Reasoning to Decision
We’ll start by reframing what “modern VLMs” are actually good at—especially where classic approaches break:
- Why video + time changes the problem (and why frame-by-frame logic fails)
- When to use reasoning VLMs vs. conventional CV pipelines
- Practical capability vs. cost tradeoffs in “thinking” models
Reasoning VLMs (and What “Thinking” Really Means)
Get a clear, engineer-focused understanding of:
- Visual chain-of-thought concepts (latent vs. explicit reasoning)
- How different model families behave in practice
- How to choose between Qwen3-VL (workhorse), GLM-4.1V (reasoning contrast), and LLaVA-NeXT (architecture/efficiency)
Video VLMs and Temporal Reasoning
You’ll learn and implement techniques to handle the hard parts of video understanding:
- Temporal grounding, event segmentation, and “what changed and when?”
- Avoiding causal hallucinations and frame-sampling pitfalls
- Building a Video Q&A demo that answers “why,” not just “what”
Agentic VLMs: Observe → Reason → Decide → Act → Remember
Go from “model outputs” to systems with behavior:
- Why agentic VLMs replace brittle rules with adaptive policies
- Persistent state, memory buffers, tool usage, and feedback loops
- Building a monitoring application that watches a video stream and decides when to raise alerts—driven by language-based decision logic, not hard-coded rules:
- detect abnormal behavior
- explain decisions
- modify behavior based on past observations (e.g., temporary exceptions via an admin instruction)
Who Should Attend
This workshop is ideal for engineers, developers, and technical leads who want to build the next generation of:
- video understanding and “what happened / why / what next” applications
- intelligent monitoring and safety systems
- adaptive, agent-like multimodal products
Prerequisites
To get the most from the hands-on work, you should be:
- comfortable with Python
- familiar with basic VLM concepts (e.g., CLIP and embeddings at a high level)
- interested in reasoning, temporal understanding, and agentic systems
Why Attend?
Multimodal AI is rapidly shifting from static perception to temporal reasoning and autonomous decision-making. This course helps you keep pace with where VLMs are going—while staying grounded in implementable techniques, real tradeoffs, and working applications you can build on after the session.
Register Today
Registration is $495. Don’t miss this opportunity to enhance your skills and stay at the forefront of computer vision technology. Register today to secure your spot in this transformative training session. (You can save $50 if you also register for the Embedded Vision Summit!)
-
-
Get in Touch
Want to contact us?
Use the small blue chat widget in the lower right-hand corner of your screen, or the form linked below.STAY CONNECTED
Follow us on X and LinkedIn.
