regl-cnn
AI
neural network demo
SHOW HN2016
How HN took it
2 positive, 7 neutral or asking, 4 mixed, 23 critical, out of 36 comments about the project. Thread score -0.52.
Execution: 2 for, 21 againstIdea: 0 for, 2 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Interesting proof of concept
From 1 comment
Pretty cool. I realize it is mainly a proof-of-concept, but decided to try out variations of scribbles[1]. Does the code make explicit assumptions about orientation, or did the…
eatbitseveryday on HN
Objections
Misreads 7s written with a dash, 1s with a tail, and 9s as 3s
From 3 comments
When I handwrite a 7, it has a dash through the middle. Everyone I know does this. Your app doesn't like these; it sees a 3.
imadfy on HNI draw my 1's the way it's often displayed on screen, with a little tail at the top and a line at the base. Your system keeps detecting this as a 2. I've also had 3's detected as…
jvdl on HNI have trouble with the 9's being returned as 3's
Sander_Marechal on HN
Returns 0 or errors on some GPUs due to glReadPixels on float textures
From 3 comments
I have tried this. It is always returning 0.
amitmerchant on HNGetting the same issue with chrome 52, windows 10 and a r9 fury. [.Offscreen-For-WebGL-043C4318]GL ERROR :GL_INVALID_OPERATION : glReadPixels: demo:1 Issue with amd cards maybe?
eksrow on HNYeah, reading from floating point textures with `glReadPixels` is not really supported on some cards or browsers, it seems.
erkaman on HN
Training data (MNIST) limits accuracy on real handwriting
From 1 comment
Yeah, the MNIST dataset is not that great it seems. At least for recognising real handwritten digits.
erkaman on HN
WebGL inference is likely slower than CPU because of data transfer
From 1 comment
The demo uses WebGL, and if you can't get the demo to work, you can here(https://github.com/Erkaman/regl-cnn) find a recorded gif that shows what it is supposed to look like.…
erkaman on HN
Requests
Automatic porting from TensorFlow to JavaScript
From 1 comment
Are there any plans to make the porting automatic with some sort of compile to JS?
rawnlq on HN
Training in the browser
From 1 comment
Since the network has already been trained, I can't imagine using WebGL is worthwhile. It would interesting to see, either, drawing samples from the equilibrium distribution or…
DanWaterworth on HN
Questions
How does the script read GPU output back to JavaScript?
From 1 comment
Intriguing! I was under the illusion that webGL was emit-only, and that Javascript programs can't read back any output generated by WebGL. I must be wrong or out of date :) So…
willvarfar on HN
Does the model assume digit orientation?
From 1 comment
Pretty cool. I realize it is mainly a proof-of-concept, but decided to try out variations of scribbles[1]. Does the code make explicit assumptions about orientation, or did the…
eatbitseveryday on HN
Alternatives named
DeepBeliefSDK and its weblas port
From 1 comment
Very well done, but not the first: 2 years ago Jetpac released DeepBeliefSDK (https://github.com/jetpacapp/DeepBeliefSDK). They were acquired by Google and development stopped.…
kcimc on HN
gpu.js
From 1 comment
Very well done, but not the first: 2 years ago Jetpac released DeepBeliefSDK (https://github.com/jetpacapp/DeepBeliefSDK). They were acquired by Google and development stopped.…
kcimc on HN
mxnet.js
From 1 comment
Nice work! I've also tried my hands at use WebGL to do deep learning in the browser (e.g. https://github.com/scienceai/neocortex/blob/master/src/lib/webgl/matmul-webgl.js). The…
transcranial on HN