photopea / unn.js

Deep Learning in JS. Alternative to TensorFlow and ConvNetJS, that is 4x faster.

Photopea.com is a free online tool for editing raster and vector graphics with support for PSD, AI, and Sketch files.

Since Photopea is not fully open-source, this repository serves as a place for bug reports, feature requests, and general discussion.

---

Supported Formats

• Complex: PSD, AI, INDD, XCF, Sketch, XD, FIG, KRI, CLIP, PXD, PXZ, CDR, UFO, Fireworks PNG, AFPHOTO, SVG, EPS, PDF, PDN, WMF, EMF.
• Raster: PNG (APNG), JPG, GIF, WebP, ICO, ICNS, BMP, AVIF, HEIC, JXL, PPM/PGM/PBM, TIFF, DDS, IFF, ANIM, TGA.
• Raw: DNG, NEF, CR2, CR3, ARW, RW2, RAF, ORF, GPR, 3FR, FFF.
• Animated: GIF, APNG, MP4, WEBM, MKV (each frame becomes a layer, can be exported back to GIF or MP4).

Where You Can Find Us

• ✉️ [email protected]
• 👍 facebook.com/photopea
• 🐦 twitter.com/photopeacom
• 📙 reddit.com/r/photopea
• 🎉 photopea.com

Other

• API, Blog, Learn, Translations

unn.js's Issues

Is it possible to make Random Ferns like classification architecture with UNN ?

Hi,

I am looking for kind of network architecture which can be trained fast in browser to match image patch of size(32x32px/16x16px) just like Random Ferns is it possible with UNN.js?

Random outputs (predictions)

Hi,

Sadly your script doesn't seems to be consistence in output (prediction).
Just run it few times (press F5 refresh Chrome console and you will see different results).
Every aprox. 5+ time it's random, not stable, doesn't matter how many iterations while training.

Run it yourself.
`

<script type="text/javascript" src="/js/UNN.js"></script> <script type="text/javascript" src="/js/UNN.util.js"></script> <script> var net = UNN.Create([ ["inpt","line",2,1,1], ["full","sigm",2,1,1], ["full","sigm",1,1,1] ], 0.5); var O = []; var In = [[0,0],[0,1],[1,0],[1,1]], Ou = [[0],[1],[1],[0]]; var prm = { method:"sgd", batch_size:1 }; for(var i=0; i< 50000; i++){ UNN.Train(net,In,Ou,prm); } UNN.GetOutput(net, [0,0], O); console.log(O[2][0]); //should be aprox 0.0xxxx UNN.GetOutput(net, [1,0], O); console.log(O[2][0]); // should be aprox 0.9xxxx UNN.GetOutput(net, [0,1], O); console.log(O[2][0]); // should be aprox 0.9xxxx </script>`

FIREFOX:

CHROME:

IE EDGE

Add license

Add some license.

Publish on NPM

You should publish this on NPM. If you for some reason can't do it, I can do it myself.

training a net for Euclidean color distance

Hi @photopea,

i'm on the lookout for js micro-libs and there are not many in the NN category. brain.js is huge, as is convnetjs and others. UNN looks interesting. i'm new to neural nets but would like to try training a NN to approximate color distance for my color quantization library: https://github.com/leeoniya/RgbQuant.js

so far i've tried a basic feed-forward neural net using https://github.com/wpmed92/backpropaganda but the results were not too good (they seemed to mostly match a guassian curve). i'm not sure if my generated training set is too small or my network topology is wrong. i don't have much intuition into what the correct parameters and training set size should be. the general advice i've heard is the hidden layer size should be half-way between the input and output layer sizes. previous discussion here: wpmed92/backpropaganda#13

maybe you would you be interested in sharing some of your wisdom or advice?

thanks!
Leon

Doesnt seem to parse from convnet

I am using

var net = UNN.util.fromCNJS(JSON.parse(mod));
alert(net);

to make a network from a cnjs string;
the mod is a string of the JSON but the code doesn't run after that.
net doesn't alert. Some more useful info might be that the model is
pretty big. Another useful thing to know is that it works when I use convnetjs.

Demo?

I'm not sure how to use this otherwise?

Does it have depthwise convolution? How would I add a new layer type if it doesn't?