Forum

VL.Lobe

hey there,

stumbled upon Lobe, a tool from Microsoft that allows you to easily train machine learning models. the idea is that you give it pictures, label them, and the software gives you a trained model you can use to classify your images.

kinda like RunwayML does, the app exposes a local HTTP server with a rest API you can send your images to, and get classification results from.

this VL plugin contains a small Classifier node that does just that : encodes your input image to base64, makes the appropriate query to the Lobe local server and parses the results in an intelligible way for your patching pleasure.

here’s a small gif showing the result I got :

lobe2
fantastische qualität gif

what’s cool here is that I just gave lobe pictures of me making 👍 and ✌️, labelled them (“One” and “Two”), and that’s it. the app can even tell you the error rate for each class so you can adjust things accurately.

this tool seems pretty recent, and for now it “just” does classification. they claim it’ll also do object detection and data classification in the future, let’s see!

you can test this yourself by installing the nuget

nuget install VL.Lobe -pre

or look at the repo here:

cheeeeers!

11 Likes

Wow this Lobe thing is great ! amazing finding
I would love to find a way to open the model in a native way ( wthout using lobe API ) or export the model as YoloV3