Hacker News Reader

Comment by raverbashing | Hacker News Reader

Counting elements from pictures is how old again now? 20 years? 30 years?

Sure, it's more reliable and you can run it on a phone. But maybe this could have been solved 20 years ago. (Yes, it might not have been as convenient, but for something that apparently is so important, it could have been applied earlier)

Japan, a country which in the 80s was synonymous with technological innovation seems to have slowed down dramatically in adopting new technologies

Edit: of course I'm not saying it's exclusive to Japan. My comment was more in the way of "the first time I saw an actual application of CV counting things in a practical setting was 20 years ago"

View on HN

If you think old, unoptimized, high-human labor business processes exist only in Japan, have I got a bridge to sell to you.

Optimizing 100 hours per year to 20 hours isn't much either..

It's 3 days, for that to break even development costs have to very low.

Of course, selling it as a generic app for counting things is a great way to do this.

That part of the article confused me too, I think it's later implied they pay that 100 vs 20 hour cost much more often than once a year.

It does mentions three months reduction of labour cost, which is a lot more than 80 hours.

People outside IT field generally don't even know how counting thing from picture is very trivial computer vision task. This is common everywhere in the world.

Are there any alternatives? Or how one should approach making an app from scratch?

The old method used to be template matching, and it probably still works well enough for problems like this. There are almost certainly now better approaches possible using Machine Learning.

In my experience, the main challenges with problems like this are dealing with varying lighting, scales, orientation and perspective. These can quickly become of diminishing returns – especially if the solution is provided as an app that provides near-immediate feedback.

My bet is they just invested a few months making a varied dataset, a few grand on mturk using labelme, some image augmentation, a few engineering tricks for a nice UX,some strategies for getting the model to converge on tricky test images, and got it to hit 100% accuracy fairly quickly.

We are working on this product in-house for 7 years now. We are an internal startup from a company that does custom computer vision and ML software.

No mturk, no tricks, a lot of work in ML and in UX as it is not a very straightforward task.

> It is not a very straightfoward task.

I'd well believe it. I used to write computer vision applications for semiconductor manufacturing equipment and there we were able to strictly control the distance from camera to object, lighting etc. and even still getting necessary reliability was not simple. When a failure could lead to damaging a whole wafer, i.e. hundreds of thousands of dollars, 99% accurate is not good enough.

It would be very questionable to use this in production, especially as pearls are very glossy and costly, without the error matrix being provided. It would seem more reliable to use the camera as a blocked / non-blocked sensor with what would amount to a cheap coin sorter.

For some reason it's very common here on HN to trivialize the difficulty of automating tasks. But you're not new here so you probably are familiar with it!

Wasn't trying to trivialise your work! Just offering some ideas of the basic ways this could be tackled. Sorry if you thought I was trivialising it, the intent was more to support the idea that modern DL can still provide solutions with relatively simple methods.

> Japan, a country which in the 80s was synonymous with technological innovation seems to have slowed down dramatically in adopting new technologies

Japan is not a country I associate with a drop in technology adoption, have you been there? It's like going to the future

Reply on news.ycombinator.com