We need your help now
Support from readers like you keeps The Journal open.
You are visiting us because we have something you value. Independent, unbiased news that tells the truth. Advertising revenue goes some way to support our mission, but this year it has not been enough.
If you've seen value in our reporting, please contribute what you can, so we can continue to produce accurate and meaningful journalism. For everyone who needs it.
Sign in. It’s quick, free and it’s up to you.
An account is an optional way to support the work we do. Find out more.
By continuing, you are indicating that you accept our Terms of Service and Privacy Policy.
WHENEVER YOU SEARCH for an image, you’re normally relying on the descriptions for an image, but new developments in image-recognition software might make it easier to find the type of images you’re looking for.
A collaboration between two teams of researchers from Google and Stanford University is creating software that will help describe the scene happening in an image, instead of just individual objects.
The software teaches itself to recognise and identify entire scenes and describe it in terms that anyone could understand.
How the software accomplishes this is by using two neutral networks. The first one deals with image recognition while the second deals with natural language processing. By using computer learning, which sees it being fed a number of captioned images and learning how the sentences provided relate to what the images show.
The developments could make it easier to group and search for the billions of images and hours of videos that are available online. Currently, Google and other sites rely on written descriptions accompanying an image to figure out what it contains, but this method is able to recognise and describe them without human assistance.
That said, it’s not perfect by any stretch of the imagination. While it’s capable of being accurate, the examples provided below show that there’s still a lot of work to be done before it’s ready for the world, either making minor mistakes or getting it wrong entirely.
According to the New York Times, the two research teams said they expect to see significant increases in accuracy as they improve their software and train these programs with larger sets of annotated images.
Still, the speed in which image recognition is improving is picking up and perhaps in the near future, we will be able to upload an image or video and it will recognise what’s happening.
To embed this post, copy the code below on your site
Test it on porn, the answers should be entertaining:
… plumber with cheesy moustache attempts to fix washing machine
…frustrated housewife now plays “hide the salami” with plumber…
… plumber and housewife get stuck playing play leap frog…ctrl..alt..delete…
Wow kind of creepy. Wonder will it be able to caption “this idiot is taking a selfie” lol.
Woulda been handy in the leaving cert German oral.
The end is nigh.
No.
Enda’s Nidge.
Wow – that’ll come in handy for when I don’t have eyes.
A red motorcycle. ..must go to specs savers
Also a funny looking skateboard.
Pretty sure the dog in the top right isn’t jumping and isn’t catching a frisbee
have your say