danott

Yes! We have several language-specific models we train against to see which performs the best for your data. Which language (or languages) do you need? (feel free to email me at dan at nyckel dot com)

Thanks for the feedback! We actually had a feature for “what is the accuracy if I only consider >80% confident samples” but we iterated away from it because people found it complicated. We’ll definitely bring it back when we can make it simple enough.

We’ve also found that people can get lost in the filters; in particular the “Not assigned” annotation filter we probably need to remove for people who have annotated all of their data.

In terms of separating training / test data: we use cross-validation so that we can abstract away the concept of train vs. test vs. validate sets.

Actually, all these examples are straight from our userbase. I can't speak to how successful they're being in their own businesses, but they seem happy with the ML. The 'barcodeless scanner' in particular is about scanning bulk foods that don't have barcodes, and it seems to work for them.

Yep. The pricing model does basically break down for model export, but I think there's a solution there. Or, said another way, if we could make it really easy to do then there's an adjacent business we could move into.

In terms of keeping customers as they grow, our view (hope?) is that these models will be continually updated because of new annotations on their end, and from new training techniques on ours. And that concept of continuous improvement will push people toward a SaaS model.

When you say chokepoint, are you referring to cost, or latency, or something else?

Well there’s a continuum of offerings, some having lots of custom control of the training pipeline and on the other side things like CustomVision that try to make it easy / hide the complexity. We consider ourselves even further to the “hide complexity” side, since we try every domain automatically vs. making you choose, re-train automatically, etc. Initial feedback from our users is that even recall vs. precision is something they don't want to have to think about. In addition, we don't limit ourselves to only vision - we'd like to be the one stop shop for ML as a service.

Oh man, we literally just got new images from our new designer and I think they went a different direction with that one :)

Haha yea I'm to blame for the crappy UI... I believe in principle with shipping things you're uncomfortable with, but it also hurts my soul.

We do think "model export" is important, but we're still getting our heads around how to do it in the most non-ML-expert friendly way. We don't think the persona we're building for wants a weights file dropped in their lap. What output / format would be ideal from your perspective?

Thanks. They're both on the roadmap; in fact we've got a handful of users in a private beta for object detection; let me know if you're interested in getting in on that!