MyCaffe

Welcome to MyCaffe - The Deep Learning solution for Windows C# Developers! Now supporting: CUDA 11.8.0, cuDNN 8.8.0 (default) Note (Compute 5.3 and higher required for __half sized memory)

CUDA 12.2.1, cuDNN 8.9.4
Note (Compute 5.3 and higher required for __half sized memory)
To download CUDA 12.2 support, see https://www.signalpop.com/products
(requires NVIDIA driver 536.67 or later)

The MyCaffe NuGet package contains the binary release of the open source MyCaffe project at http://www.github.com/mycaffe.

This NuGet package contains the MyCaffe control, MyCaffe image database, MyCaffe basecode, MyCaffe extras, MyCaffe trainers and the MyCaffe gyms, and now has full online help available at https://www.signalpop.com/help.

To get started, please see the MyCaffe Programming Guide at https://www.signalpop.com/mycaffe/ and check out the samples at https://github.com/MyCaffe/MyCaffe-Samples.

IMPORTANT

MyCaffe now uses the .NET Framework 4.8 throughout.

MyCaffe is 64-bit software, so in order to use it, you MUST make sure the 'Prefer 32-bit' build option is UN-CHECKED. To un-check this option, right click on your project, select 'Properties', select the 'Build' tab, and un-check the 'Prefer 32-bit' option.

Some of the MyCaffe Samples noted below expect that you have already installed the MyCaffe Test Application and loaded both the MNIST and CIFAR datasets into your SQL EXPRESS database. To learn more about this, please see the MyCaffe Test Application at https://github.com/MyCaffe/MyCaffe/releases.

When building a .NET Framework Console Application, you should first remove the following sample files that are automatically added to your project:

MyCaffeSample.cs FormMyCaffeSamples.cs WebCamDialog.cs AleSample.cs OnnxSample.cs

These files use Windows Forms and are only intended to show you how to use MyCaffe and its various parts but are not needed to use MyCaffe.

REQUIRED SOFTWARE:

1.) Download and install the latest MyCaffe Test Application from https://github.com/MyCaffe/MyCaffe/releases (includes CUDA and cuDNN distributable DLL's)

• or -

1.a) Install NVIDIA CUDA 11.8.0 which you can download from https://developer.nvidia.com/cuda-downloads

1.b) Install NVIDIA cuDNN 8.8.0 which you can download from https://developer.nvidia.com/cudnn

(To download MyCaffe Support Installs for CUDA 12.0 or CUDA 11.7, please see https://www.signalpop.com/portal)

• next -

3.) Download and install Microsoft SQL Express 2016 (or later) from https://www.microsoft.com/en-us/sql-server/sql-server-downloads, IMPORTANT: You must install the full SQL or SQL Express from the link above which is different that the 'lightweight' version of SQL installed with Visual Studio.

FINAL INSTALLATION STEPS:

1.) Register the 'CudaControl.dll' by running the following in a CMD window with Administrative privileges: a.) Go to the 'packages\CudaControl\nativeBinaries\x64' directory b.) run 'regsvr32 CudaControl.dll' to register.

2.) Register the 'AleControl.dll' by running the following in a CMD window with Administrative privileges: a.) Go to the 'packages\AleControl\nativeBinaries\x64' directory b.) run 'regsvr32 AleControl.dll' to register.

3.) Download and install the MyCaffe Test Application located at https://github.com/MyCaffe/MyCaffe/releases. Before using MyCaffe, you will need to create a new MyCaffe database - the MyCaffe Test Application will do this for you automatically. In addition, you can use the MyCaffe Test Application to load the MNIST dataset into your MyCaffe database.

4.) A sample dialog 'FormMyCaffeSamples' and file 'MyCaffeSample.cs' have been added to your project. These files demonstrate how to create and use MyCaffe to solve several common tasks such as the following:

To run the sample, just add the following code to your 64-bit C# project:

    MyCaffeSampleUI.FormMyCaffeSamples dlg = new MyCaffeSampleUI.FormMyCaffeSamples();
    dlg.ShowDialog();

This sample contains the following demonstrations:

-- CLASSIFICATION --

• LeNet - train a classification problem by learning the MNIST dataset with the 'LeNet' model.

• SiameseNet - train a classification problem by learning the MNIST dataset with the 'Siamese Net' which feeds positive and negative images against an anchor to learn a distance separation between classes by learning with two parallel networks.

• TripletNet - train a classification problem by learning the MNIST dataset with the 'Triplet Net' which feeds anchor, positive and negative images simultaneously to learn a distance separation between classes by learning with three parallel networks.

-- RECURRENT LEARNING --

• CharNet - train the recurrent CharNet to learn how to write a Shakespeare sonnet.

-- REINFORCEMENT LEARNING --

• Policy Gradient - train a policy gradient reinforcement learning model to learn how to balance the pole in the Cart Pole gym.

• Policy Gradient - train a policy gradient reinforcement learning model to learn how to play and beat the ATARI game Pong.

• DQN - train a DQN reinforcement learning model to learn how to play and beat the ATARI game Breakout.

-- NEURAL STYLE TRANSFER --

• Neural Style Transfer - learn the artistic style of Vincent van Gogh and paint a photograph in his artistic style to produce a new piece of art.

-- CONTROL TESTS --

• ONNX - test the ONNX model format.

• ALE - test the ALE ATARI Game Simulator.

• WebCam - test the WebCam control used to input video data.

-- SEQ2SEQ with ATTENTION

For examples of SEQ2SEQ with attention, please see the Seq2SeqChatBot MyCaffe-Sample at https://github.com/MyCaffe/MyCaffe-Samples/tree/master/Seq2Seq

Each sample sends their output and status to the Visual Studio output windows, so when first running these samples, you may want to run them in Debug mode.

To use any of the samples, just call one of the methods listed above, build with the 'Prefer 32-bit' Build Setting as UNCHECKED, build and run.

Follow the instructions in this file to get started using MyCaffe!

NOTE: Make sure to build in x64! Uncheck 'Prefer 32-bit' in your Build settings or errors will occur.

For more information on the MyCaffe implementation of Policy Gradient Reinforcement Learning, see MyCaffe: A Complete C# Re-Write of Caffe with Reinforcement Learning by D. Brown, 2018.

If you use MyCaffe in your application, feel free to also use either of the 'Powered by MyCaffe' logos added to your project to show that you are using MyCaffe!

MyCaffe uses Microsoft SQL (including the free Microsoft SQL Express) for data storage. To create the MyCaffe database, please install the MyCaffe Test App which can be found at https://github.com/MyCaffe/MyCaffe/releases.

In addition to database management, the MyCaffe Test App provides tools to automatically load the MNIST and CIFAR-10 datasets, and provides access to over 1500 automated tests.

If you are looking for tools to visualize and debug your deep learning models, check out the SignalPop AI Designer, where you can visually edit models, manage training sessions and debug your training in real-time. The SignalPop AI Designer now supports deep auto-encoder networks, domain adversarial neural networks (DANN), policy gradient reinforcement learning, recurrent LSTM learning (using cuDNN), Neural Style Transfer, Siamese Nets and Triplet Nets for metric learning.

If you would like to help improve MyCaffe with your input, add a pull request or comment to the open-source project at: https://github.com/MyCaffe/MyCaffe

You can try the SignalPop AI Designer demo version for free at: https://www.signalpop.com/products

For more information on innovative Deep Learning AI development tools that use the MyCaffe AI Platform, such as the SignalPop Visual AI Designer, see https://www.signalpop.com.