Not only is this not computer vision, but the image processing employed is quite naive and the results are poor. One need only move the Simplify slider all the way to the right on the default N/S America globe icon and then start inching it back to the left to see a solid example of this. I'm aware that I'm being very dismissive here, but the "it uses computer vision" marketing bs really grinds my gears.
I am curious to know, does the HN software have an automatic trigger for comments that mention "title" or is it more a case of mods looking at the front page every now and then and if a comment about the title is sufficiently highly voted in a frontpaged post you look into it and make a decision?
The latter. I used to have an automatic trigger but it broke when I ported my client software from one obscure Lisp to another, and I haven't yet gotten it working again.
typically you would leave the large sizes as-is, the filters are mostly useful for the smaller sizes where you don't see the diamond kernel effect.
not sure why people so upset about the name, mathematical morphology is commonly used in computer vision. As far as marketing BS goes it's pretty tame compared to some of the stuff I've seen on hn.
Loops are commonly used in computer vision. And yet here we are not calling loops computer vision.
> not sure why people so upset about the name
Learned people get upset when they see grandiose or otherwise erroneous claims about trivial things, regardless of the field, because it signals either bumbling naiveté or intent to deceive, and both are perceived as dangers to the community.
Central America is there. It does kind of drop out of the reduced images when you get down towards 48x48 but that makes sense because it's the smallest feature.
As for the random button the author replied and they've locked it to a single image for now to help with the load on the servers. Single image is easier to cache the responses for.
Can't get it to work on Firefox due to it trying to load jQuery from code.jquery.com where the CORS-policy disallows it. Specifically:
> Cross-Origin Request Blocked: The Same Origin Policy disallows reading the remote resource at http://code.jquery.com/jquery-1.12.4.min.js. (Reason: CORS header 'Access-Control-Allow-Origin' missing).
This is pretty standard image processing techniques, which you can streamline yourself. Not sure if there exists out-of-the-box solutions, but you can easily create your own with for example scikit-image (python).
This is incredibly useful, regardless of the veracity of its "computer vision."
A technique that seems to work well for it with simple drawing style icons is to thicken and simplify as the image gets smaller, so it seems like the stroke of the drawing is consistent across sizes.
really cool idea but it didn't work for me. It distorted the logo and made it look much worse. Would be great if you generated a unique url for each icon that is uploaded so i could share it with you.
The techniques detailed on the page (reweighting lines, removing fine details) seem like a worthwhile reason to at least use a different file for the "tiny" image. Unless SVG renderers start using those techniques at a small enough size.
This is for favicons, where svg support is pretty spotty.
Even with svg, they ultimately become rasters once they are on the screen. So for favicons it might still be beneficial to control the scaling yourself with a tool like this.
Genuine question: Are people creating app icons while they're on the go? Or am I the only one who defers all development to one's laptop (or desktop) computer.
Agreed, I only work on my desktop, my phone is for not essential tasks. But we might be the minority, I’ve seen people use their tablets for Everything
The page loads, but normally i see earth logo on desktop, but on phone it's just blank. This is on android chrome. Looks to be a issue with the margins/alignment of the image, i think.
I'm going to guess it's from a legacy system and the code was copy/pasted when creating that new site... but will be interested to hear the final reason.
These kind of fundamental methods in computer vision are commonly called "image processing".
"Computer vision" typically refers to computations that have an interpretative component (image analysis), i.e. programs that actually "see" something. Interpretation basically refers to the inference of the state of the "world" (or a part thereof) from an image. Image processing is often used in pre-processing for computer vision applications, but it also has many other purposes which do not need image analysis, such as in photo and video editing, as well as in generative applications, e.g. 3D engines and video games (which can in some sense be seen as the inverse of vision). Therefore, image processing is not exactly a subset of computer vision.
When
I visit the site for the first time
Then
I expect to see something related to the title from the HN submission
Except
I see a blank page and an app that has broken in a fundamental way that I do not have expertise/time to look into
It looks like it's just doing server-side http://www.imagemagick.org/Usage/morphology/#basic
Hello massive diamond kernels.