Animas-V

The vision half of Animas — a neural-network and object-detection engine in pure C. Detects, trains and streams on your own hardware, with the network cable out.

Schedule a Discovery Meeting

Vision

What it is

Detection you can watch happen, on the box, disconnected.

Animas-V is QBO's object-detection engine: the original Darknet, cleaned up, made memory-clean and brought up to the current generation of YOLO models — all of it still in C. No framework, no Python, no model server to stand up alongside it.

It is not a demo harness around someone else's weights. Training runs in the same engine, on the same box, so a new class can be taught in front of you and then found in the next frame.

Inference and training in one engine, on one machine.

What it does

Detects on live video

Point it at the bundled feed, an RTSP camera, a webcam or a file. Frame count and frame rate tick on the page as it runs.

CPU to GPU, one click

Move the same inference between CPU and GPU while it runs, so the frame-rate change is something you watch rather than something we claim.

Learns a new class, live

Train a class from scratch in front of you, then watch it find the thing it just learned. Training, not just inference, runs on the same box.

The modern YOLO ladder

Current-generation detection, not a decade-old model with a new coat of paint — the accuracy you would expect from a modern stack, without the stack.

Streams at hardware speed

H.264 over WebRTC with NVENC hardware encoding, measured side by side against a plain JPEG path — resolution, frame rate, kilobytes per frame, bandwidth and loss counted for both.

Evaluates honestly

Your own classes trained end to end, with dataset preparation and mAP@0.5 evaluation in the same tool.

Why it is built this way

One binary, no stack

Detection, training, evaluation and the web viewer are the same C program on one port. Nothing to integrate.

Memory-clean, proven

Memory-clean and warning-free, proven by the test suite on every change — not asserted in a datasheet.

Straight to the hardware

CUDA and NVENC run against the physical driver stack at full speed — no hypervisor, no emulation layer in between.

Know what the link costs

Encoder, bandwidth and loss are measured in the page, so you can see what a stream costs before you put it on a constrained link.

See it detect. Then teach it something new.

See it alongside Animas-R and the QBO API.

Request a demo→