Google Lens knows more about what’s in your photos than you do
- Thread starter JournalBot
- Start date
Hopefully their image recognition also gets a lot better. I've tried searching through photos using keywords and the results it returns are poor at best.
Upvote
-4
(10
/
-14)
This would be super cool if it works well. Identifying birds/plants just by pulling up your camera would replace quite a few identification apps. Pulling something like that off well would be quite the hurdle, though.
Upvote
40
(41
/
-1)
Im sure a robot that can learn from its environment would love context. Think personal healthcare. The doctor can look at the picture from your phone with context to bring his attention to where it should be in relation to healthcare
Upvote
7
(7
/
0)
My worry with Google is always when is this coming and will it actually be here a year from now...It seems like they have been advertising stuff like this for years but it never has been very useful due to low accuracy and needing to pull up special apps to do it. Plus I saw a lot of cool demos in last IO that never seemed to be implemented.
Upvote
20
(22
/
-2)
If you're taking a picture of your spouse or kids, then sure, there isn't much utility. If you're on vacation and pointing the camera at a statue in Sofia, Bulgaria, and Google Lens is able to tell you who it is and some of the history behind it, well, this is very useful. It's basically what Google Glass did, but less dorky for everyday use, especially when travelling to places you may be completely unfamiliar with.
It will depend on GPS coordinates and such, yes, but if it is at least 80-85% accurate, it's really a great tool to have.
Upvote
47
(49
/
-2)
Funny, I was thinking just the other day why Google didn't make Google Goggles into this. I can definitely see the utility. Diagnose that boil, identify that beetle, find out where that spare part belongs etc. Maybe in combination with a relaunched Glass (if they can figure out how to negate the creep factor)?
Upvote
12
(12
/
0)
I imagine recent advancements in neural networks means their system's ability to identify things has improved enough for it to be a viable feature.
Upvote
13
(13
/
0)
My experience has been the opposite. The face recognition is excellent and it's easy to find photos of our kids or kids with Mom etc. I also had good mileage finding photos of my son learning to cycle or holidays at certain location or years. It's not yet like searching in Gmail but it's vastly better than what I had to do 10 years ago. Creating a bat mitzvah slideshow is now a lot easier.
Best of all, it improves over time and I no longer have to install and maintain gallery on php or all the other software I had used over many years.
Integration with drive means that I am never locked in and always take my photos somewhere else. It's a brilliant platform.
Upvote
15
(15
/
0)
Off the top of my head..
Take a picture of a weed in your lawn and have it identified including remediation recommendations.
Take a picture of a leaf and find out what kind of tree or plant it is.
Upvote
29
(29
/
0)
While bringing up info on an item is nice, I'd like it if the photos got tags or something so I can easily search on my phone or tablet for pictures rather than scrolling through them manually.
Associate date, time, location and the various content in the photo.
That is useful.
Associate date, time, location and the various content in the photo.
That is useful.
Upvote
4
(4
/
0)
Oh, great, another autoconnect-to-wifi feature that is impossible to exploit in any way, I'm sure...
Meanwhile, on planet Earth, my 'built-in' AT&T wifi feature (aka autodownloaded to my Nexus by Google after inserting sim chip) autoconnects to every AT&T Wifi, which Android then warns me isn't secure, prompting me to turn off wifi entirely, dinging me on data.
Grape job, guys!
Meanwhile, on planet Earth, my 'built-in' AT&T wifi feature (aka autodownloaded to my Nexus by Google after inserting sim chip) autoconnects to every AT&T Wifi, which Android then warns me isn't secure, prompting me to turn off wifi entirely, dinging me on data.
Grape job, guys!
Upvote
-17
(5
/
-22)
Of course there is substantial and obvious utility for the end-user here. Without it, data collection stuff like this wouldn't work because no one would use it.
Upvote
3
(5
/
-2)
I remember the morning that Google Now on my Nexus 5X recommended that I search for cat pics after I had taken pictures of cats the night before.
That's either a f--king convenient coincidence or they've been beta-testing this for a while now.
That's either a f--king convenient coincidence or they've been beta-testing this for a while now.
Upvote
9
(9
/
0)
This has been around since 2010 with an app called Word Lens. Google acquired the company behind that in 2014.
Upvote
15
(15
/
0)
Would love to see something like this for identifying the various different skin disorders such as non threatening skin blemishes versus the various types of skin cancers especially melanoma(s?).
Upvote
4
(5
/
-1)
Upvote
9
(9
/
0)
Most important identification tests for me (that I think I will use a lot):
1) poison ivy (and other plants that I register as covering me in hives)
2) spiders, bugs, snakes (if deadly, add number of minutes till death to visual timer and auto call emergency services, else just tell me how to treat it)
3) Amazon one click ordering for anything I take a picture of (along with other sellers' prices as metadata).
I can think of lots of uses of this...
1) poison ivy (and other plants that I register as covering me in hives)
2) spiders, bugs, snakes (if deadly, add number of minutes till death to visual timer and auto call emergency services, else just tell me how to treat it)
3) Amazon one click ordering for anything I take a picture of (along with other sellers' prices as metadata).
I can think of lots of uses of this...
Upvote
11
(12
/
-1)
I would love that. Birds are fairly easy to identify, given that there aren't a million of them, and none of them are poisonous. However, I'd love to be able to point the camera at a plant, and have it tell me whether it's hemlock or not.
Upvote
0
(1
/
-1)
didn't they already have Google Goggles app which can identify object in photo, but the image recognition are sketchy.
also did you all forget about the debacle when a photo of an African Americans man get recognized by Google as gorilla in the past, in the end they can't fix the flaw and rather get the gorilla tag removed.
so i doubt it can identify more than human do.
also did you all forget about the debacle when a photo of an African Americans man get recognized by Google as gorilla in the past, in the end they can't fix the flaw and rather get the gorilla tag removed.
so i doubt it can identify more than human do.
Upvote
-1
(1
/
-2)
Seriously. Up to now it's been impossible to get into wi-fi networks that list their access ID and password on a sticker affixed to the router. Now it'll be a piece of cake. Come on, Google!
Upvote
-3
(1
/
-4)
How long till someone uses this to identify a mushroom they picked, the mushroom gets mistakenly labelled as edible, the person eats it and gets extremely unhealthy or worse, google gets sued?
Not to say this tech isn't cool though.
Not to say this tech isn't cool though.
Upvote
8
(8
/
0)
As a person with a botany background who identifies a lot of plants, I have learned that the internet is not always the best place for plant identification. There are many sites where folks load up misidentified photos and those false identification gets propagated on down the line. I have found that it is very difficult to get someone to "unlearn" what they found on the internet.
The USDA plant database is quite good and has a procedure in place to correct misidentified plants. Hopefully Google will have similar features. The "Duck Test" doesn't always work well for plants. If it looks like a rose and smells like a rose, it isn't necessarily a rose.
Upvote
17
(17
/
0)
Someone mentioned statues earlier, another mentioned leaves, etc.; I've come across murals here in NYC that I had no idea who was in them, including one where I was fortunate enough to ask the artist right there who it was. If this technology works for murals, paintings, etc., I will really love this technology.
I'm also kind of hesitant about what kind of data profile this will allow Google to build (they're not offering this out of the kindness of their heart, after all) for users. But at this point, unless you want to live in a van down by the river...
I'm also kind of hesitant about what kind of data profile this will allow Google to build (they're not offering this out of the kindness of their heart, after all) for users. But at this point, unless you want to live in a van down by the river...
Upvote
5
(5
/
0)
You seem to be conflating a manual-step saving convenience feature with a bullshit carrier-imposed threat vector.
If I have to open my camera and point it at the SSID and password info, I'm hard-pressed to see the sneaky attack vector. I suppose there's the possibility of character spoofing, where the SSID looks like one word but is actually made of special characters, but if they are similar enough to fool my eyes, I don't see how the camera would do much better. In any event, I have to open the camera and point it at something for this to kick in.
Upvote
8
(8
/
0)
I like using secure passwords, but I also hate having to navigate through three keyboards to get the mix of characters, numbers, and symbols required to type in a password on my phone. I really do not understand how this idea isn't an improvement on the current experience.
Also it will make things so much easier when crashing on a friend's couch after a night out. So much easier than typing "monkeybananawifi" and "aR6^[jE2xL" on a phone keyboard when drunk.
Upvote
6
(7
/
-1)
Can it identify trees from their leafs? I've been looking for a good app for Android to do that since I got my first Android phone back in the 2.3 days. iOS users have Leafsnap but the best Android app is Virginia Tech Tree ID which requires a bit too much human guesswork to be super accurate.
Upvote
3
(3
/
0)
The auto recognition is very cool, but in terms of data collection, it means Google is analyzing the content of what you are looking at, and reading and analyzing all the text too. That's a big advancement in their ability to understand you, feeding that data to a deep analysis that learns a lot more than people expect. Almost any kind of information can be determined from photos.
Upvote
5
(6
/
-1)
this pretty much works already (for date, location metadata) in gphotos.
What they're doing is improving the content recognition.
I have a lot of pictures of chinese artwork and artifacts. Today, gphotos only does 50/50 on finding pictures featuring dragons or tigers. I'd love it if it got better at that.
It already does a good job of finding pictures of my family. And an ok job of finding pictures of my dog. There was a bear in a tree in our yard recently and gphotos couldn't come up with the pictures I took when I asked for "bears" but it found them when I asked by date. And it filtered down to a correct subset when I added the "tree" term. It just didn't recognize the bear's head peering down from the tree.
Upvote
2
(2
/
0)
What you do then though is leaking the SSID and password to Google, since all the photo analysis doesn't run on your phone but on Google's servers.
All of this is just a window into what Google (not your phone) knows and learns about you and your data.
Upvote
6
(7
/
-1)
Or the bugs that are ravaging my okra. I'm willing to sacrifice that, but if they pose a threat to my sweet peppers, I'm willing to violate Geneva, Hague and the CWC.
Upvote
7
(7
/
0)
[url=https://arstechnica.co.uk/civis/viewtopic.php?p=33342939#p33342939:1yr3a1ps said:
Er, yes, actually it sounds really useful. There was a similar thing they had a while ago called goggles, but it never worked brilliantly. If they've got the tech ready for prime time and it's actually accurate now, that would be fantastic. Personally it would be useful for identification of plants and birds but I'm sure most people have things they'd occasionally like to identify, and being able to do it from a photo would be awesome.
Thing is, I wonder whether they have the level of detail yet. It's one thing to be able to say "that's a mountain" and another to be able to say "that's the Grandes Jorasses and the prominent feature in the middle is the Walker Spur", or "that's not just a screw, it's a 3.5x40mm single thread screw with a PZ2 head". The more detail they can eke out of their machine learning, the better.
Upvote
2
(2
/
0)
I'm sure this will be useful once perfected. However I can't help but think about the adverse effect high quality intelligent image processing will have on our privacy (if it isn't the final nail in the coffin). I really have to get stricter with acquaintances uploading pictures of me and mine to the internet.
Upvote
1
(1
/
0)
- Status