Where is the Model Zoo

[kilickaya]

I wonder if anyone converted pre-trained Caffe (or any else) models like AlexNet, VGG, HYBRID to ConvNet.js format?

If not, how could it be possible?

[karpathy]

there wouldn't be, you can't use a VGGNet in Javascript :) Maybe one day...

[marcj]

Mh, what is the reason for this?

[kilickaya]

Probably due to memory issues (A bowser is allowed to only use 500MB memory..)

[codinfox]

and models like that will take way too long for a js script to evaluate, not to mention to train.

[UniqueFool]

and models like that will take way too long for a js script to evaluate, not to mention to train.

we were recently talking about this in the context of the Brain.js and synaptic frameworks. The idea now is to support a different back-end that supports GPU-level parallelization via OpenCL using the arrayfire-js bindings.

Obviously, that will not help with the browser use-case, but the discussion is still interesting: https://github.com/arrayfire/arrayfire-js/issues/10

The only way to make OpenCL work in the browser (without WebCL) is doing something like this: http://composition.al/blog/2015/02/24/to-opencl-from-javascript-via-js-ctypes-or-how-we-rewrote-the-river-trail-firefox-extension/

They basically came up with a firefox extension to do this.