Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
tcwicks
@tcwicks
at Tensorflow.Framework.smart_module.smart_constant_value(Tensor pred)
tcwicks
@tcwicks
Basically the framework is a multi threaded / distributed training framework. I'm trying to create one example brain + agent so that how it works is clear. For the example brain I'm trying to create a basic style GAN coupled with a quantized variational encoder
The reason I need tf.cond is to illustrate how to dynamically turn on / off brain functions. Essentially a composite brain with two functions. VAE encoder decoder and GAN.
Haiping
@Oceania2018
Do you have any UnitTest that I can work with you together?
Your project is too big. I can’t catch up. Please provide a small piece that debuggable.
Lukas Donkers
@Lunariz
Hi there! I'm trying out Keras.net and just had my model blow up because it hit 100% accuracy 😂
Tried the TerminateOnNaN callbacks but it seems that callbacks aren't supported?
Is there any way I can make sure that the result of Fit() doesn't go from 99.9% accurate to 0.001% due to the loss being NaN at 100% accuracy?
Lukas Donkers
@Lunariz
ok, just managed to implement it as a custom callback as described here SciSharp/Keras.NET#13
Just had to find it in the keras source
Haiping
@Oceania2018

Today, I closed the TensorFlow .NET 0.1x.x series and stopped adding new features. Future code will be developed based on tensorflow 2.x. Although I know that 0.1x is very imperfect, I can only focus my limited attention on maintaining new code.

I released TenosrFlow.NET 1.15.0 just now and the corresponding full cross-platform customized version of tensorflow prebuild binary 1.15.1 CPU and GPU versions. At the same time, the existing master will be the future version 0.2x.

JeffMushmo
@JeffMushmo
Is there a timeframe/schedule in which you plan to release the 0.2x version or plan to release a version with support for Tensorflow 2.x as an official nuget package?
Kerry Jiang
@kerryjiang
everybody are curious about 2.0
Zoltán Lehóczky
@Piedone
Hey everyone! We're working on a .NET hardware accelerator: https://github.com/Lombiq/Hastlayer-SDK Do you think something in SciSharp would benefit from this? So any performance-intensive process in any of the libraries which is highly parallelized and compute or memory bound?
Haiping
@Oceania2018
@Piedone I'm very interested in your project, will take a deep look.
Haiping
@Oceania2018
So excited: TensorFlow 2.x Eager mode works: https://twitter.com/ScisharpStack/status/1236461970896928768
Zoltán Lehóczky
@Piedone
@Oceania2018 great, let me know if you have any questions. I can also do a demo with screen sharing if you'd like (to the whole SciSharp community too).
Haiping
@Oceania2018
@henon Are you interested in this Hastlayer ? Turning software into hardware. @/all Anybody wants to see the demo, we might host a web meeting if more people are interested in your project. @Piedone Here is link from your website just post again: https://www.youtube.com/watch?v=973ag7yMgK0
Meinrad Recheis
@henon
@Oceania2018, we actually used FPGA's in my company to do incredible audio processing for 64 channels of audio all at the same time
so yes, FPGAs are very powerful. but they are also hard to program
Haiping
@Oceania2018
@henon Do you have tool to make it easier, maybe @Piedone 's technology help you.
Meinrad Recheis
@henon
That would have been great to have when we needed it
instead now all our FPGA code is written in VHDL which is a horror to code
Zoltán Lehóczky
@Piedone
Being a software developer who dived pretty deep into FPGA development for Hastlayer, I agree :).
BTW at the time Hastlayer is mostly about compute acceleration, i.e. using FPGAs as accelerators for a .NET program running on a standard PC. We're evaluating how this could be used in embedded systems like drones and satellites (yes, we're at the very beginning of a .NET satellite platform!).
Haiping
@Oceania2018
@/all Any idea about this issue ? SciSharp/TensorFlow.NET#525
JamesLear92
@JamesLear92
Hi all, I've got a quick question...
I'm trying to build a neural network for chess, which would have an input shape of 8x8x10, and an output shape of 8x8x64
the problem is, I can't work out how to implement a multidimensional output & labels. All examples I have seen are flat labels
Haiping
@Oceania2018
@JamesLear92 You can add one more dim to 8x8x10 as [8, 8, 10, label].
image.png
Jonas Chan
@jonaschan
Hello, i am trying to convert from an image byte array to mat. is it possible to do so in SharpCV?
Essentially in the YoloDetector in Sci-sharp examples, i need Yolo to take in a byte array instead of reading an image from disk before passing it to the detector
I tried converting the byte[] => bitmap => ndarray => mat but it comes up with an exception saying that "Not Supported"
Jonas Chan
@jonaschan
@Oceania2018 Hi! I am trying this example: https://github.com/SciSharp/SciSharp-Stack-Examples/blob/master/src/TensorFlowNET.Examples/ObjectDetection/YoloCoco.cs. (Please look at line 94) However, instead of using SharpCV to read a file from disk, i want to be able to pass in an image byte array instead. Is there a way to do so?
Sathish Raja Bommannan
@SathishTheDna
Hi, I have a doubt. Is it possible to utilize Tensorflow.NET for inferencing inside Unity for Android Builds?
Haiping
@Oceania2018
image.png
@jonaschan
dariush
@eulerleibniz
Hi. I'm new to both SciSharp and UWP. I'm trying to use NumSharp( and basically any other SciSharp package) in my UWP app. First I need to know if it is fully supported? if Yes then here is the problem. It works perfectly in Debug mode, but when I switch to Release mode i get ILT0005 error after a long time compile and it seems in the time it wants to run the app. It's caused only when Native Compiler is used. After looking into it found out that it happens with all Nuget packages without signature. So for example I can use Microsoft.ML package which is signed b Microsoft and is Trusted by Visual Studio. Looked for SciSharp Signature but couldn't find it. How can i fix this problem? How to add SciSharp to my trusted repositories? Problem is i can't find the Certificate if any. Regards
Иван Сердюк
@oceanfish81_twitter
@jonaschan , @dcmartin could help - he has spme experience in using Yolo. He is running https://github.com/dcmartin/motion-ai
Jonas Chan
@jonaschan
@Oceania2018 I have managed to convert a byte array to ndarray to pass in into the new Mat constructor but it is saying "NOT SUPPORTED"
Haiping
@Oceania2018
image.png
@eulerleibniz SciSharp's package are signed in ML.NET's open snk.
Иван Сердюк
@oceanfish81_twitter
Can someone explain how to use SharpCV for the video stream, coming from WebRTC or ORTC ?
dariush
@eulerleibniz
@Oceania2018 Thanks for the quick response, my friend. But can u give a bit more details on how to use them? A reference would be more than enough. Regards
Arnh
@Arnh
Hi All. I have downloaded the examples on github, ma it complains miss Tensorflow.Binding sln. Any hints? thanks
Arnh
@Arnh
Hi. Someone could check the examples zip? There are some errors, miss the*.bindings.
Haiping
@Oceania2018
@Arnh Please switch to 0.15 branch.
@all It's time to pursude tensorflow team to make .net binding in the official repo. Welcome to join discussion: tensorflow/tensorflow#38027
Meinrad Recheis
@henon
good idea @Oceania2018