News for hackers. Stuff that matters.
Re:Show HN: A neural net that critiques your logo by twiss on Thursday August 31, 2017 @09:12AM (#15139992)
Cool! I tried it with our logo (https://www.airbornos.com/images/logo-mark.png), which is super generic (I spent $0 on it). It came up as 96% unique, even though there's two icons of clouds in the list of similar icons. Furthermore, it would be cool if it looked for similarity with a database of other logos as well, and not just icons.

Other small point: maybe you could show the uploaded logo again next to the other logos in the "Color/Contrast" and "Overall" categories, to (probably) drive home the point how bad your logo is in comparison.

Re:Show HN: A neural net that critiques your logo by dschep on Thursday August 31, 2017 @09:21AM (#15140057)
Yeah, two nearly identical versions (probably only resolution differences) of our logo resulted in 21% & 94% similarity scores
Re:Show HN: A neural net that critiques your logo by Jack000 on Thursday August 31, 2017 @09:28AM (#15140105)
it's definitely not perfect. I think the embedding-vector approach of image search isn't scale invariant.
Re:Show HN: A neural net that critiques your logo by dsnuh on Thursday August 31, 2017 @08:21PM (#15145179)
Forgive me if this is a naive question, as I am not well versed in ML.

Would it be possible to train using Mechanical Turk so that you get real human feedback and perception of quality?

Re:Show HN: A neural net that critiques your logo by wingerlang on Friday September 01, 2017 @01:24AM (#15146327)
Considering all the bad logos in the world that humans make, use and like - I wouldn't bet on it.