91–100 of 170 posts
It's only a matter of time before a repeat of Microsoft's last AI experiment (Tay), when the Internet teaches CaptionBot all of the positions in the Kama Sutra.
Edit: recognizes Stalin though http://i.imgur.com/9W6wqUd.png
As far as I know this was the first research to do the super cool thing to combine multiple neural nets trained on different data in super cool ways:
"Now, what if we replaced that first RNN and its input words with a deep Convolutional Neural Network (CNN) trained to classify objects in images? Normally, the CNN’s last layer is used in a final Softmax among known classes of objects, assigning a probability that each object might be in the image. But if we remove that final layer, we can instead feed the CNN’s rich encoding of the image into a RNN designed to produce phrases. We can then train the whole system directly on images and their captions, so it maximizes the likelihood that descriptions it produces best match the training descriptions for each image."
AND
"Our alignment model is based on a novel combination of Convolutional Neural Networks over image regions, bidirectional Recurrent Neural Networks over sentences, and a structured objective that aligns the two modalities through a multimodal embedding"
Ohh I got a good one: "I am not really confident, but I think it's a close up of a plane with a blue umbrella." http://imgur.com/FYucrda
Earlier quoted context omitted.
Hah, yeah but when I got it working again, this time with a screenshot of Jules from Pulp Fiction, it thinks his gun is a camera. > I am not really confident, but I think it's a man holding a camera. Source Image: http://www.cinemablend.com/images/news_img/79237/pulp_fictio...
Imagine this tech matures and it can be incorporated with bodycams for police, when confronting a subject with objects in their hands it may be able to confidently estimate the probability of being a firearm or not, with better predictability than the police/people.
While we're here, let's go the full way and set up a proveable and public way to train a robocop, and I'd trust that more than a human cop. The awkward moment when AIs have more brains than cops (at least under the US system).
It's only a matter of time before a repeat of Microsoft's last AI experiment (Tay), when the Internet teaches CaptionBot all of the positions in the Kama Sutra.
I don't mean to offend, but I'm left wondering if the creators of image recognition services disincentivize their neural nets from recognizing something as an ape, gorilla or chimpanzee so as to avoid the same mistake Google made when it falsely recognized black people as gorillas [1].
[1] http://blogs.wsj.com/digits/2015/07/01/google-mistakenly-tag...
CaptionBot team here. Thanks for the images and captions! Please keep sharing them and give us feedback.
Wondering if you plan to open up a caption API of any sort? Can definitely use something like this. If you desire the training feedback, then that could be added as well as part of the API. I'd be willing to do that for some images. So if you do add a training feedback API, please make it optional.
Pretty impressive - gave it a few profile photos and it did suprisingly well, correctly identifying "A couple walking on a beach at sunset," "a man looking out a window", etc. It struggled with wildlife photos - a pack of arctic wolves was "a sheep standing in the snow", and penguins swimming was "a bird flying over a body of water" (close but no cigar).
