TensorFlow Lite for mobile developers (Google I/O '18)

[@elenacairo7736]

Good morning nice to meet you here I'm watching here conception tarlac Philippines

[@danilof79]

Hello! Can you tell us where to get pc app that you use for retraining model to recognize tensorflow logo?

[@paulstanway9478]

Hi, great talk where can I find the MS Windows inference / train demo app featured at 29:48s ?

[@mesutsimsek35]

Respect from Turkey, thank you Google

[@mikexue5104]

do you have a compatible(verified) mobile device list for TF Lite? I can't run TFLite gesture detection on my Huawei Mate9, it's tooooo slow. after some searching work, I found out those Huawei phones with NPU chipset maybe good for TFLite.

[@sohaibarif2835]

I think it would be better to ask this on Github but on the off chance someone knows here; it looks like the tensorflow.contrib.lite.Interperter does not exist in Python tensorflow1.8 and neither does convert_savedmodel. A similar thing happened when I was trying some code demoed on the tensorflow youtube channel in the 1.7 release(latest one at the time) but worked when 1.8 was released. So is this demo for 1.9?

[@manfyegoh]

any chance for react native user?

[@Daniel23544]

Thanks for the talk. Can we get rid of the boisterous birds for I/O 2019?

[@pennhwan7363]

its awesome ,can i read that retrained model on github?

[@deletesobaka]

Cool, but when are you going to provide official TF builds for Raspberry Pi?

[@sohaibarif2835]

I have been planning on buying a pi 3. Quick question: does the prebuilt Python tensorflow not work on PI on Ubuntu?

[@deletesobaka]

Sohaib Arif google provide builds just for x86_64 linux systems, since they target mostly hi-end systems wuth Nvidia gpus. So you have either to build it from source yourself or install a wheel built by someone else. The pitfall in building it yourself is that arm architecture is not supported for bazel, the built system you need to build TF, so you have to patch it first, find correct versions of java, bazel and tf that will work nicely together, spend several hours of pure building time which may randomly crash due to lack of ram or whatever. So, for now, just find one of the projects on github who provide prebuilt wheels and go from there.

[@pauldacus4590]

Am I hearing birds chirping in the background?

[@JinkProject]

yes. if you look close enough, you'll see them flying all over the place.