Hacker News Reader

Comment by rhizome | Hacker News Reader

What's that old joke? Something like in the 1980s a Media Lab teacher gives the class a computer vision assignment where they're supposed to be able to tell whether or not an image contains a bird, and 40 years later they're still working on it? Lobe.ai reminds me of trying to identify plants with Google Goggles 10(ish?) years ago. It didn't work very well then, and then Google killed Goggles. Side note: none of the "click on the leaf feature" web-based plant identifiers gave a satisfactory answer either.

View on HN

Google Lens (on the stock Android camera app) works pretty great for identifying plants to me. I'd say it's right about 95% of the time (in New England) if I can see a flower, and about 75% of the time for a leaf.

I've used Google Lens a little bit, to see if it could be used for object identification.

It always seemes assume that I'm trying to buy something and tries to find products me that are somehow visually related to what it's looking at, or else to the content of some lettering that it's able to detect.

So I gave up on it.

Are there some settings I can tweak?

Unsure. That maybe sounds like the Google Lens standalone app from a couple years ago. The Google Lens mode with the stock camera app never shows me anything to buy. ....actually, I just pointed it my rollerball pen, and it identified the pen correctly and then showed me a link to the pen on Alibaba. But I think that was just the image it was matching. It never does that for natural objects.

That explains it; it is the Google Lens app I used. Or does it?

However, if you go to the Google Store page for it, it does tout this feature:

"IDENTIFY PLANTS & ANIMALS Find out what that plant is in your friend's apartment, or what kind of dog you saw in the park."

I have the current version, which was updated on August 13, 2020.

I will give it another spin, and also look for the lens mode in the camera app. (My phone is from Google, so the camera is the Google one; if any Android camera app has a Google Lens mode, it should be that one.)

I was going to say, Lens works incredibly well. Working with a farm box of random vegetables that we couldn't identify, Lens had a 95% accuracy and instantly, too.

"Working" can be interpreted differently. 75% accuracy (?) is not that great for predicting a specific class (depending on distribution on course).

If you need high accuracy, let's say for e.g. estimating eco system performance based on specific plant distribution, 75% is very low (especially if you want to feed it into another predictor) compared to a professional field biologist.

That's not a joke, a visual object/pattern recognition system was a summer project at MIT in 1966 under Seymour Papert. They didn't get very far.

https://dspace.mit.edu/handle/1721.1/6125

Ah I should have accounted for that, I was a bit flip because IME it has become a quip since then. I didn't mean to minimize it.

That's awesome they have the assignment for download.

Seek (the iOS) app works great -- I use it all the time on walks to identify plants (and even bugs and birds, if I can get close enough and they'll sit still). I think it's based on iNaturalist, which has a large community of professional botanists etc. collecting and classifying image data.

I think your comment is behind the state of the art. I bought a house with a beautiful garden in the backyard. I didn't know what any of this stuff was or how to take care of it. The app my wife got (not sure what it is) got all of them correct except 1, with a single photo. That's for at least 50 different plant species.

We're really excited to bring machine learning to more people and make it more accessible. What we are most excited about is empowering people to create custom machine learning models for specific and personal use-cases and solve problems in new and unique ways

I believe it is this xkcd https://xkcd.com/1425/ sorry research teams :D

Ah ha, that must be what I'm thinking of in calling it a joke.:)

I recall that in the 18th century people tried to build a computer, so... What's your point? Computer vision didn't work back than, and it does work now, but not for everything. If you don't understand it you should go read about it, not posting pointless replies

There are a number of new apps that take a crack at it. I've been using PlantSnap, and it seems to do OK. It doesn't always get it, and I wouldn't stake my life on it, but it's been good for assuaging idle curiosity.

There is an app called Picture This which is very good at identifying plants from flowers / leaves.

Reply on news.ycombinator.com