Building a computer vision model from scratch sounds straight forward—until you actually do it.
Draw boxes. Write code. Tune models.
Then lighting changes—and you're retraining. Again.
Dr. Paulina Varshavskaya @GroundlightAI broke it down:
➡️ Counting parts on a moving conveyor belt.
Doable—but painful.
There’s a better way.
With Groundlight AI:
🟣 24/7 human-in-the-loop annotation
🟣 AutoML: models trained tuned for you
🟣 ML Ops: seamless retraining & deployment
Production-grade CV—no ML experience needed.
▶️ code.groundlight.ai
Want to automatically count objects on a conveyor belt? See how you can build this computer vision application yourself - Github repo here: github.com/groundlight/conve…
Create a computer vision model to count objects using simple natural language prompts (ex: "Label each package in the image"). A sneak peak from Groundlight's dashboard:
Automatically count 🔢 parts that are moving on a conveyor belt.
⚡️ Connect any IP camera or image stream to the Groundlight API
⚡️ Point the camera to the objects that are moving on the conveyor belt
⚡️ Install Groundlight Edge to ensure rapid object counting
⚡️ Use a natural language query such as "Count the number of packages on the conveyor"
⚡️ Handle results programmatically in your Python code with one API call
Dr. Farnoosh at Groundlight AI shares a demo of counting people in the Groundlight office "wearing blue 👕 t-shirts" as they walk through a doorway.
🧵see below for more...
Traditional VMS are great at generic people-counting, but they struggle 😩 with anything custom. Here, we use object detection and counting through Groundlight—a visual understanding service that handles training and labeling for you under the hood.
The code’s here if you want to take a look: github.com/groundlight/emplo…
Note: It uses the business tier of Groundlight, but if you’d like to explore the demo or the service, we'd be happy to set you up with a free access.
Groundlight is now available for use with Agentic AI as the Groundlight MCP Server. You can build a sophisticated video stream monitoring computer vision application complete with alerts by having a conversation with Claude Desktop or your favorite AI coding agent. groundlight.ai/blog/building…#mcp#agentic
👀 Ever wonder how Vision Language Models (VLMs) actually work?
We wrote a technical guide breaking down:
– VLMs vs LLMs
– How images are tokenized into embeddings (just like text!)
– How an LLM is combined with a vision-specific model to work together for visual understanding
– Diagrams a live ViT tokenizer you can test
We made an open-source MCP server github.com/groundlight/mcp-v… that turns HuggingFace zero-shot object detection pipelines into tools that Claude and others can use to locate objects or zoom (crop) to an object. Conceptually vision capabilities as tools are complementary to VLM's reasoning powers. In practice the zoom tool allows Claude to see small details much better. More on our approach in the blog post: groundlight.ai/blog/vision-a…. We're working on extending mcp-vision's capabilities and welcome community contributions.
#ComputerVision#MCP#modelcontextprotocol#AI#LLMs#VLMs#MachineLearning
🚨Today’s the Day! Live at 12 PM PST 🚨
Learn how the powerful combination of @GroundlightAI and @Universal_Robots automates machine tending - you're not going to want to miss this 45-minute webinar!
🚀 New from Groundlight: Object Counting 🔢
We’re excited to announce a major step forward in simplifying computer vision — the launch of our new Counting Mode, for accurate, real-time object counting.
Whether you're counting cars 🏎️ , people 👯, or pallets, our latest feature empowers teams to count the exact things they want to count — no training data or ML expertise required.
This is a game-changer for industries like:
📦 Logistics & Warehousing
🏭 Manufacturing
🚗 Traffic & Parking Analytics
🏬 Retail Operations
And more!
Huge kudos to the Groundlight team for continuing to make computer vision accessible to every developer. Counting mode is available to users on the Business plan - try it out and let us know what you think!
👉 Read the full press release: businesswire.com/news/home/2…#VisualAI#ComputerVision#ObjectCounting#AIforDevelopers#EdgeAI#GroundlightAI