Korso · Products
Helios
Computer vision you can describe in plain language.
Helios lets you describe what to look for and see labeled boxes on a live camera feed. It runs locally on a CUDA GPU. This is a demo for exploring the interaction, not a deployable monitoring product.
Demo video coming soon
Capabilities
- Describe what to findUse ordinary language to describe a concept; there is no fixed class list or retraining step.
- Choose the detection modePresence, PPE, and danger-zone modes are selected by the presenter and never inferred automatically.
- Count what is in viewSee current counts and running totals as objects pass through the frame.
- Mark an area on the feedDraw a region directly on the video and flag detections inside it.
- Keep the result readableState is evaluated over time so a flag holds instead of flickering.
- Configure per machineKeep host-specific tuning in configuration rather than changing application code.
How it works
- Describe the requestA model turns the request into concrete concepts once per query.
- Run detection locallyThe detector finds those concepts on each frame on your GPU. There is no cloud vision loop.
- Return an annotated feedThe server draws the boxes and returns an annotated frame to the browser.