Sketching Interfaces

As with anything: there is a time and place. Design systems are wildly powerful for scaling, creating a cohesive language across a brand or product, and moving quickly.

Just because you have a design system doesn't mean the buck stops there. You can keep building new roads if you know where it is you need to go that the existing roads won't take you.

1: “Notes on the Synthesis of Form” by Christopher Alexander.

If they can actually pull this off it I can see it being very useful. And if it works really well in conjunction with other experiments currently ongoing, I can imagine it even affecting the way teams are structured and how a designer's day to day work goes.

Step 0: you should have some training data. Draw it < I believe 50 of each class you want to detect should be enough to start, but you can think about doubling it>

First you need look at the opencv library, specifically in the findcontours function https://docs.opencv.org/2.4/modules/imgproc/doc/structural_a... - I'm linking to the c++ api but the python has the exact same function.

Make sure your contours are clean of rubbish data because this can affect the outcome of the classifier.

With the extracted attributes of each individual class you pass on this data to a multi class classifier (http://scikit-learn.org/stable/modules/multiclass.html)

Parameterize your model, Split your data into training and testing ... yada yada ...

With the generated model, you can receive individual detected contours and it will give you the fitting probability to each class, use it to feed an api which provides data to a react component that renders html.

>> ps: Ah you need an image capture app which reads the image in real time, pass the filters and detect contours to then pass it to the classifier.

>> ps2: I'm available for projects :P

- also www.computervision.ai enabled me to get started with images files pretty fast, drag and drop to train a classifier