Teach a computer to see what you see
Label your photos by clicking on things. Press Train. Get a model that spots them for you, running on your own machine. No Python, no cloud, and no machine learning background needed.
Windows, macOS and Linux. Free account, no card.
A recreation of RedenFlow labelling crates on a warehouse shelf, then training a model whose accuracy climbs to 94 percent.
Three steps, and none of them involve code
This is the whole product. There is no stage where you are expected to open a terminal, install a dependency, or find out what a tensor is.
- Step 1
Point at the thing you care about
Draw a box, or click once and let it trace the outline for you. No file formats, no folder structure to learn, no annotation schema to design.
- Step 2
Press Train and walk away
One number tells you whether it is working. When it stops improving, it stops. You never choose an optimiser or read a loss curve unless you want to.
- Step 3
Get something you can actually run
A trained model, and a small script that runs it on a folder of photos or a camera. Not a notebook. Something you can hand to somebody else.
This is it, actually running
Sixteen photographs of a fruit line, labelled by clicking, trained on a MacBook Air, and then counting oranges on its own. No cloud, no notebook.
Your photos never leave your computer
There is no upload step and no cloud storage. RedenFlow talks to us to check your licence and for nothing else.
The labelling is the part that is fast
Click an object and it finds the edges. Most of the time you are confirming rather than drawing, and that is where the hours go.
It runs on the machine you have
A graphics card makes training minutes instead of hours, and RedenFlow tells you on the first run whether yours is being used.
The part where most tools stop
Plenty of things will help you label pictures and train a model. Far fewer will hand you something that runs on the machine in the room where the problem is. RedenFlow asks four questions and writes out a folder you copy across.
Onto the device you name
A PC or industrial computer, an NVIDIA Jetson, or a Raspberry Pi. On the last two it builds the fast version of the model on the device itself, because a TensorRT engine only loads on the chip that made it.
Reading the camera you have
A USB camera, a network camera over RTSP, a video file, or a folder of pictures. No vendor SDK, no capture card.
Doing the job you picked
Show a live window, count what goes past, write a spreadsheet, save an annotated video. Counting uses tracking, so one thing crossing the line is one, not forty.
No Docker image to build, no server to stand up, no notebook. What the wizard writes, in detail.
Find out in an afternoon whether this solves your problem
The free plan is not a trial that expires. It is the whole application with smaller limits, so you can label a few dozen photos and train something real before deciding anything.
Download RedenFlow