7 ms·

Author here. Happy to answer any questions, hear feedback.

by yantrams 8y ago

sansnomme 8y ago
Traditional algorithms or ML? Image vectors?
- yantrams 8y ago
  Traditional algorithms and image vectors. I used a conconction of existing region and contour based techniques and threw in some original ideas as well.
  - sansnomme 8y ago
    Elaborate on some of your ideas or do a blog post?
    - yantrams 8y ago
      Been meaning to do a blog post on this forever now :/ I will do it at some point for sure at http://linkdot.link http://linkdot.link This is a great starting point in case you are interested in knowing more -- http://www.staff.science.uu.nl/~kreve101/asci/smi2001.pdf http://www.staff.science.uu.nl/~kreve101/asci/smi2001.pdf More recently, I've been exploring some ideas of Tversky.
- Raphael 8y ago
  One way is to scale the images down and threshold to black and white. At smaller scale they should be identical or much closer.
pizza 8y ago
It would be cool if you could draw an icon instead of uploading/linking one, like how Google Translate lets you draw characters
- yantrams 8y ago
  Great idea. I'll definitely look into it. Thanks for the suggestion. Cheers
peterburkimsher 8y ago
Nice idea! I did a similar but different project for 32x32 icons. https://iconpush.github.io/ https://iconpush.github.io/ Please help yourself to my icon data - I spent a while collecting it and hope it can be useful to someone!
- yantrams 8y ago
  Thanks for the share! These icons are golden! I did something similar for logos that could be of interest to you - http://compute.vision/brands/colorpicker.html http://compute.vision/brands/colorpicker.html
soneca 8y ago
I spent some time confused by the animated running horse and how it was supposed to be related with the other icons. Until I found out that it was the loading gif. I would change there for something more standard.
- dessant 8y ago
  I found the running horse to be an adoring touch. The author's site is also worth checking out: http://linkdot.link http://linkdot.link https://en.wikipedia.org/wiki/Sallie_Gardner_at_a_Gallop https://en.wikipedia.org/wiki/Sallie_Gardner_at_a_Gallop
  - soneca 8y ago
    I believe if it would be both adorable and more easily understandable if it was a smaller image, giving a visual hint that is fulfills the role of an icon, not a featured image
- Normal_gaussian 8y ago
  It was the other way for me, I linked it to the duckduckgo icon svg and the horse started. I thought it was a loading animation (and took more than a few seconds) so I threw it to another monitor and continued reading HN. ...30 minutes later the horse is still running and I'm like 'wtf? what does a horse have to do with the DDG logo?' close tab. read comments... It turns out the app doesn't handle svg (it is actually in the to do list) and returned a 500, but the failure was never presented to the user.
  - yantrams 8y ago
    Ouch. Sorry to have put you through that :/ I will fix it asap. And I agree with the parent too. Maybe some text below the horse that says fetching results... would make things clearer.
    - Normal_gaussian 8y ago
      haha, no bother, we all know whats up here; I'm just busy bikeshedding. Massive kudos for delivering; I like it a lot :D
slig 8y ago
Hi, how did you source the whole noun project icon collection?
- yantrams 8y ago
  I scraped it from their website and then asked for their permission by sharing this link with them. They appreciated that I linked all the icons to their website and gave their consent to make this public.
  - slig 8y ago
    Ah, I see. I was hoping there was an archive to download somewhere. Thanks!
amelius 8y ago
How do/would you measure the accuracy of the algorithm?
- yantrams 8y ago
  The MPEG-7 dataset[0] is what most researchers use to benchmark shape similarity algorithms. There are couple of other datasets that I used that I can't recollect now. These datasets are relatively simple with a single shape as opposed to logos, icons that comprise multiple elements in different configurations. I would test on the MPEG-7 dataset to begin with and once the precision and recall values are good enough go ahead with testing on logos and icons. I must've manually tested the algorithm more than a 100,000 times probably because that was the only way to do with untagged datasets. Quite tedious indeed. This version gives out pretty decent results about 7-8 out of 10 times I'd say. [0] - http://www.dabi.temple.edu/~shape/MPEG7/dataset.html http://www.dabi.temple.edu/~shape/MPEG7/dataset.html

Author here. Happy to answer any questions, hear feedback.

sansnomme 8y ago

Traditional algorithms or ML? Image vectors?

yantrams 8y ago
Traditional algorithms and image vectors. I used a conconction of existing region and contour based techniques and threw in some original ideas as well.
- sansnomme 8y ago
  Elaborate on some of your ideas or do a blog post?
  - yantrams 8y ago
    Been meaning to do a blog post on this forever now :/ I will do it at some point for sure at http://linkdot.link http://linkdot.link This is a great starting point in case you are interested in knowing more -- http://www.staff.science.uu.nl/~kreve101/asci/smi2001.pdf http://www.staff.science.uu.nl/~kreve101/asci/smi2001.pdf More recently, I've been exploring some ideas of Tversky.
Raphael 8y ago
One way is to scale the images down and threshold to black and white. At smaller scale they should be identical or much closer.

pizza 8y ago

It would be cool if you could draw an icon instead of uploading/linking one, like how Google Translate lets you draw characters

yantrams 8y ago
Great idea. I'll definitely look into it. Thanks for the suggestion. Cheers

peterburkimsher 8y ago

Nice idea! I did a similar but different project for 32x32 icons. https://iconpush.github.io/ https://iconpush.github.io/ Please help yourself to my icon data - I spent a while collecting it and hope it can be useful to someone!

yantrams 8y ago
Thanks for the share! These icons are golden! I did something similar for logos that could be of interest to you - http://compute.vision/brands/colorpicker.html http://compute.vision/brands/colorpicker.html

soneca 8y ago

I spent some time confused by the animated running horse and how it was supposed to be related with the other icons. Until I found out that it was the loading gif. I would change there for something more standard.

dessant 8y ago
I found the running horse to be an adoring touch. The author's site is also worth checking out: http://linkdot.link http://linkdot.link https://en.wikipedia.org/wiki/Sallie_Gardner_at_a_Gallop https://en.wikipedia.org/wiki/Sallie_Gardner_at_a_Gallop
- soneca 8y ago
  I believe if it would be both adorable and more easily understandable if it was a smaller image, giving a visual hint that is fulfills the role of an icon, not a featured image
Normal_gaussian 8y ago
It was the other way for me, I linked it to the duckduckgo icon svg and the horse started. I thought it was a loading animation (and took more than a few seconds) so I threw it to another monitor and continued reading HN. ...30 minutes later the horse is still running and I'm like 'wtf? what does a horse have to do with the DDG logo?' close tab. read comments... It turns out the app doesn't handle svg (it is actually in the to do list) and returned a 500, but the failure was never presented to the user.
- yantrams 8y ago
  Ouch. Sorry to have put you through that :/ I will fix it asap. And I agree with the parent too. Maybe some text below the horse that says fetching results... would make things clearer.
  - Normal_gaussian 8y ago
    haha, no bother, we all know whats up here; I'm just busy bikeshedding. Massive kudos for delivering; I like it a lot :D

slig 8y ago

Hi, how did you source the whole noun project icon collection?

yantrams 8y ago
I scraped it from their website and then asked for their permission by sharing this link with them. They appreciated that I linked all the icons to their website and gave their consent to make this public.
- slig 8y ago
  Ah, I see. I was hoping there was an archive to download somewhere. Thanks!

amelius 8y ago

How do/would you measure the accuracy of the algorithm?

yantrams 8y ago
The MPEG-7 dataset[0] is what most researchers use to benchmark shape similarity algorithms. There are couple of other datasets that I used that I can't recollect now. These datasets are relatively simple with a single shape as opposed to logos, icons that comprise multiple elements in different configurations. I would test on the MPEG-7 dataset to begin with and once the precision and recall values are good enough go ahead with testing on logos and icons. I must've manually tested the algorithm more than a 100,000 times probably because that was the only way to do with untagged datasets. Quite tedious indeed. This version gives out pretty decent results about 7-8 out of 10 times I'd say. [0] - http://www.dabi.temple.edu/~shape/MPEG7/dataset.html http://www.dabi.temple.edu/~shape/MPEG7/dataset.html