Jack000

44 ms·

451.
by Jack000 9y ago
there's another project that does this: https://erikbern.com/2016/01/21/analyzing-50k-fonts-using-de... the results are a bit blurry but as the author notes the next step is to use a better loss function
452.
by Jack000 9y ago
keep in mind S3 bandwidth is roughly 20~30x more expensive than a VPS. I put my photo/video blog on S3 reasoning that it'll handle spikes better. At one point it reached the front page of HN and cost me $450. Afterward I looked at
453.
by Jack000 9y ago
I suspect the tough part of whatthefont isn't so much the visual search but having a large database of labelled fonts. There are other services that does this but personally I've always heard good things about whatthefont.
454.
by Jack000 9y ago
it uses VGG16 to extract font features, then compares those features to figure out the best contrast between fonts. feel free to use the vectors in the Github.
455.
Show HN: Choosing font combinations with deep learning(github.com)
121 points by Jack000 9y ago|18 comments
456.
by Jack000 9y ago
The problem with talking about housing prices in Vancouver is that it's a heavily politicized issue with all sides pushing an agenda. this article specifically compares jan 2016 to jan 2017 in order to push the narrative the tax worked
457.
by Jack000 9y ago
In 2014 the CAD was nearly at parity with USD, now it's 0.74 The average price of goods has increased accordingly. As far as I can tell the actual value of a house has stayed about the same, and the purchasing power of Canadians has go
458.
by Jack000 10y ago
I've always thought that was an air gap - if you lived in a world where your ghost could be hacked you wouldn't want to risk being plugged in just for some text input.
459.
by Jack000 10y ago
haven't been to India but anecdotally the amount of public defecation between DT SF and rural China is about the same - ie. quite rare in practice but brought up ad nauseam in online discussions, often to serve a political point. on th
460.
by Jack000 10y ago
for the stars shot the next obvious step would be to automatically segment the image with optical flow. Could also try solving the hand-held problem with ORB feature matching. given the previous posts from the googleblog I kind of expected
461.
by Jack000 10y ago
China too. Reddit users seem tolerant until you find yourself in the outgroup. HN is the new reddit for me.
462.
by Jack000 10y ago
I wonder what the closest to "YC classic" is in 2017. I was going to apply to YC fellowship but disappointed to see it's cancelled.
463.
by Jack000 10y ago
maybe prior to colonialism. The modern conservative Chinese narrative is more of an inferiority complex rather than the opposite. It's easy to see why the CCP is against religion - the party is rooted with the cult of Mao, which is a d
464.
by Jack000 10y ago
that's a great question! The main reason is that pix2pix works at a set resolution by default, and changing it is non-trivial. I'm actually not sure how to structure the Unet for a 5x1 image - if you have any ideas let me know (I&
465.
by Jack000 10y ago
will it be fully 3d printed (no metal/carbon tubes)? I think the main problem with hobby robot arms is flex in the arm itself and backlash in the motors. Low cost arms (eg. dobot) have "0.1mm repeatability" with no load but t
466.
by Jack000 10y ago
very very cool. One thing I noticed that with image to image translation tasks the output tends to be a bit "organic" looking, like the photo to map example. With photographic output it's not noticeable, but it's very ja
467.
by Jack000 10y ago
it's possible, but the main challenge would be to make a sufficient sample of 10-color palettes to train on. I am making something that disambiguates between foreground and background, but still 5 colors for now.
468.
by Jack000 10y ago
GANs are good at generating plausible images given a set of examples, which is exactly what we're doing here. A statistical approach might work, but you'd be essentially interpolating your existing samples. A GAN is capable of gen
469.
by Jack000 10y ago
the choice of a CNN makes more sense if you think of a color palette as an extremely low-resolution image rather than a basket of colors. positioning in a palette matters because colors are perceived relatively.
470.
by Jack000 10y ago
the GAN is actually just pix2pix: https://github.com/phillipi/pix2pix I just resized the convolutional layers to 1px height since the vertical dimension isn't that relevant in this case. I do have a few bash scrip
471.
by Jack000 10y ago
the issue I see is that most sites have a white/beige background, which could get repetitive for the purposes of design inspiration. Maybe train a separate model for bootstrap/direct insertion.
472.
by Jack000 10y ago
I think color theory alone isn't enough to produce good color palettes. eg. try going to https://color.adobe.com and choose one of the color rules, then go to https://color.adobe.com/explore and compare wit
473.
by Jack000 10y ago
currently the colors don't encode context, so you have to be a bit creative in applying them. I am working on something that does encode context though.
474.
by Jack000 10y ago
the model isn't totally deterministic, but aside from that the main palette is actually an ensemble of 12 separate models trained on slightly different hyper-parameters. Colors are kind of subjective, and this is just meant to add some
475.
by Jack000 10y ago
sure, you could just use the same API the main website does, it's unauthenticated: curl 'http://colormind.io/api/' --data-binary '{"model":"default"}' curl 'http:&
476.
by Jack000 10y ago
yeah definitely, it would just have to train on a (relatively large) sample of such palettes. I'll look into it!
477.
by Jack000 10y ago
The torch7 models need to be loaded in memory, which is the main bottle neck. I've got about 30 or so trained models on disk and they take turns on the homepage. Maybe there's a better way to do this that I'm not aware of tho
478.
Show HN: Colormind – Color schemes via Generative Adversarial Networks(colormind.io)
399 points by Jack000 10y ago|52 comments
479.
by Jack000 10y ago
I'll be there! Throw up the "Y" hand sign if you're from HN, we can have a little hn meetup.
480.
by Jack000 10y ago
maybe I'm missing something, but PLA is just 3d printer filament. There's only two plastics commonly used for extruder-style 3d printers, and they're ABS and PLA.

More ›