Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Neural Artistic Captions (toronto.edu)
46 points by tim_sw on Oct 5, 2015 | hide | past | favorite | 10 comments


Editorialized:

> I've never had a chance to escape

> No matter what happens to me, I am sure of it, and I am still here.

> Ahead of course, I did not see a sign that he was going to stop me from passing over the city. I avoided eye contact with the local police station, and once again I scanned the rooftops for signs of stopping. No matter how much damage I've done [snip] the main gate is still wide open. The signs grew louder and louder with each passing word.

It's either been exposed to lots of adventure stories involving great escapes and existential issues or that's just outright creepy.


Kafka-esque


This is extremely cool. An algorithm that writes stories about a photo. It understands (some) English. Not enough to be completely sensible, but way more than markov chain generated stuff. And the output is definitely related to the photos.

I wish there was some more details about this. All I could find was this tweet: https://twitter.com/rsalakhu/status/650860432341778432

>Many existing models generate boring captions. These captions are biased by the style of adventure books


I went up to the parent dir of the URL and found this link: http://kelvinxu.github.io/projects/capgen.html


Even though it doesn't make much sense, I enjoyed the unintentional pun in the story of the bathroom picture. "The toilet flush with the toilet s surface caused him to sink deeper into the murky water."


Yes enjoyable, would love to know the training data. It seems the phase 'he did not know what to say' occurs quite a bit in the training data. Maybe more interesting if they pruned the comments down to the most succinct sentence or two


Can you please add some additional information for the URL you have provided?


Looks like it's the home page of Ryan Kiros at the University of Toronto. Here's a paper with his name on it with some related work, though I suspect this particular output is from a later project.

http://arxiv.org/abs/1502.03044


I'm very curious about the apparent association between elephants and caves. As another poster said, the training data would doubtless illuminate...


Very... entertaining! :)




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: