TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Subpixel: A subpixel convolutional neural network implementation with Tensorflow

210 pointsby jgoldsmithover 8 years ago

11 comments

maxanderover 8 years ago
So, basically, this is the thing in a crime detective movie where the forensic analyst is looking at a terrible pixelated surveillance camera still and says &quot;enhance,&quot; and the computer magically increases the resolution to reveal the culprit&#x27;s face.<p>Just another entry on the &quot;things that are supposed to be impossible that convolutional nets can do now.&quot;
评论 #12619935 未加载
评论 #12619656 未加载
评论 #12620642 未加载
评论 #12620258 未加载
评论 #12621668 未加载
anotheryouover 8 years ago
I think it&#x27;s always problematic to compare to images upscaled via nearest-neighbor. The big pixels are hard to parse for our brain, we detect all the blocky edges.<p>A good content unaware upscaling would be nice (one of the default photoshop algos)<p>I also wonder what they used for the downscaling. I see 4x4 pixel blocks, but also some with 3px or 7px lengths.<p>This looks pixely and is supposed to be a source file?: <a href="https:&#x2F;&#x2F;raw.githubusercontent.com&#x2F;Tetrachrome&#x2F;subpixel&#x2F;d2e28518d2ce75a1ae7dde7727603b8652d683cb&#x2F;images&#x2F;lowres_input.png" rel="nofollow">https:&#x2F;&#x2F;raw.githubusercontent.com&#x2F;Tetrachrome&#x2F;subpixel&#x2F;d2e28...</a>
评论 #12622005 未加载
Roboprogover 8 years ago
Interesting image &quot;upscale&quot; algorithm.<p>I&#x27;m not familiar enough with the field to understand how the &quot;neutral net&quot; part feeds in, other than to do parallel computation on the x-pos, y-pos, (RGB) color-type-intensity tensor interpolated&#x2F;weighted into a larger&#x2F;finer tensor.<p>(linear algebra speak for upscaling my old DVD to HD, that sort of thing)<p>At the risk of exposing my ignorance, this has nothing to do with &quot;AI&quot;, right? It&#x27;s &quot;just&quot; parallel computation?
评论 #12619712 未加载
评论 #12620690 未加载
评论 #12620763 未加载
markisusover 8 years ago
It seems that this subpixel convolution layer is equivalent to what is known in the neural net community as the &quot;deconvolution layer&quot; but it is much more memory and computation efficient. The interlacing rainbow picture was a bit hard to understand until I read this <a href="https:&#x2F;&#x2F;export.arxiv.org&#x2F;ftp&#x2F;arxiv&#x2F;papers&#x2F;1609&#x2F;1609.07009.pdf" rel="nofollow">https:&#x2F;&#x2F;export.arxiv.org&#x2F;ftp&#x2F;arxiv&#x2F;papers&#x2F;1609&#x2F;1609.07009.pd...</a>
ameliusover 8 years ago
Interesting. They should post more examples (not with just faces), or make an online demo, like waifu2x [1]<p>[1] <a href="http:&#x2F;&#x2F;waifu2x.udp.jp&#x2F;" rel="nofollow">http:&#x2F;&#x2F;waifu2x.udp.jp&#x2F;</a>
评论 #12620925 未加载
zokierover 8 years ago
I&#x27;m not sure, but there seems to be something wonky in the input images. They are very blocky, so I thought that they would be just pixel doubled (or quadrupled) from low-res pictures, but the blockiness lacks the regularity I&#x27;d expect from pixel-doubled images.<p>How were the input images prepared?
评论 #12620299 未加载
ericjangover 8 years ago
The explanation in the README of the github project is excellent and well-written! Here&#x27;s a really great set of animations by Vincent Dumoulin on how various conv operators work: <a href="https:&#x2F;&#x2F;github.com&#x2F;vdumoulin&#x2F;conv_arithmetic" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;vdumoulin&#x2F;conv_arithmetic</a>
评论 #12620312 未加载
Keyframeover 8 years ago
This is impressive! But, I&#x27;ll be really impressed once this &#x27;new thing&#x27; brings us roto masks in motion. That is, isolating objects from background on a movie with pixel-perfect accuracy. It will also make a lot of people out of job and a lot of people happy at the same time.
评论 #12620706 未加载
robertkrahn01over 8 years ago
And I always wondered how those photo enhancers in Blade Runner worked...!
thoreauwayover 8 years ago
ENHANCE. ENHANCE.
评论 #12619765 未加载
imaginenoreover 8 years ago
The problem with subpixel images is that there are RBG and GBR monitors. Not only that, there are horizontal and vertical variations. And there&#x27;s no way to tell which one the user is using on the web. And that&#x27;s not even counting all the mobile number like pentile.<p>It&#x27;s still useful though, browsers, for instance, could use it for displaying downscaled images.
评论 #12620158 未加载
评论 #12620139 未加载