Skip to content

Releases: getnamo/TensorFlow-Unreal

v0.9.0 for UE4.20

24 Jul 20:44
Compare
Choose a tag to compare

Changes

  • Tensorflow upgraded to 1.9.0
  • UnrealEnginePython plugin dependency updated to 1.6.0
  • Socketio-client-ue4 plugin dependency update to 0.9.0
  • Compile fixes for 4.20
  • TFAudioCapture module updated with: support for native delegates, fixes for full capture of audio on finish, audio level provided on each chunk.

breaking dependencies cause bump to 0.9

v0.8.0 for UE4.19

18 Mar 18:33
Compare
Choose a tag to compare

Changes

  • UnrealEnginePython plugin dependency updated to 1.5.0
  • Socketio-client-ue4 plugin dependency update to 0.7.3
  • Compile fixes for 4.19
  • breaking dependencies cause bump to 0.8

Note on GPU release
Tensorflow v1.6.0 GPU depends on CUDA v9.0 and cuDNN v7 see https://www.tensorflow.org/install/install_windows for detailed instructions on installing those pre-requisites. If you can't satisfy those requirements, use the cpu version.

v0.7.0 for UE4.18

05 Mar 14:31
Compare
Choose a tag to compare

Changes

  • Tensorflow dependency update to 1.6.0
  • UnrealEnginePython plugin dependency updated to 1.4.2, small changes to path cleanup.
  • Fixed site reload after installing pip issue #19, remove the need to restart your project after installing python dependencies
  • Refactor for upypip, the threaded pip utility. New suggested import format: import upypip as pip
  • Breaking dependencies cause bump to 0.7.0

Note on GPU release
Tensorflow v1.6.0 GPU depends on CUDA v9.0 and cuDNN v7 see https://www.tensorflow.org/install/install_windows for detailed instructions on installing those pre-requisites. If you can't satisfy those requirements, use the cpu version.

v0.6.0 for UE4.18

27 Jan 05:12
Compare
Choose a tag to compare

Changes

  • Tensorflow dependency update to 1.5.0
  • Socket.io plugin dependency updated to 0.7.1
  • UnrealEnginePython plugin dependency updated to 1.4.1, synced with upstream.
  • Embedded Python changed to 3.6.4
  • Breaking dependencies cause bump to 0.6.0.

Note on GPU release
Tensorflow v1.5.0 GPU depends on CUDA v9.0 and cuDNN v7 see https://www.tensorflow.org/install/install_windows for detailed instructions on installing those pre-requisites. If you can't satisfy those requirements, use the cpu version.

v0.5.0 for UE4.18

14 Jan 20:38
Compare
Choose a tag to compare

Changes

  • Tensorflow dependency update to 1.4.0
  • Socket.io plugin dependency updated to 0.7.0
  • UnrealEnginePython plugin dependency updated to 1.3.0
  • 4.18 compile fixes for this plugin. Breaking dependencies cause bump to 0.5.0.

Note on GPU release
Tensorflow v1.4.0 GPU depends on CUDA v8 and cuDNN v6 see https://www.tensorflow.org/install/install_windows for detailed instructions on installing those pre-requisites. If you can't satisfy those requirements, use the cpu version.

v0.4.1 for UE4.17

28 Aug 02:50
Compare
Choose a tag to compare

Changes

  • Added an optional parameter to TFPluginAPI call shouldRetrain. It syncs with TensorFlowComponent->ShouldRetrain boolean. Use this boolean to check whether you script should force retrain the network.

Note on GPU release
Tensorflow v1.3.0 GPU depends on CUDA v8 and cuDNN v6 see https://www.tensorflow.org/install/install_windows for detailed instructions on installing those pre-requisites. If you can't satisfy those requirements, use the cpu version.

v0.4.0 for UE4.17

24 Aug 12:18
Compare
Choose a tag to compare

Changes

  • Breaking API changes in TFPluginAPI. setup -> onSetup, train -> onBeginTraining, runJsonInput-> onJsonInput.
  • onSetup is now multi-threaded and called asynchronously before onTrain (if enabled) on begin play.
  • add an API on TFPluginAPI to call a game thread disptach event on TensorflowComponent with just calling callEvent('MyEventName','MyStringOrJsonData'). Makes it easy to send periodic information during training, including showing what training data is being used.

Note on GPU release
Tensorflow v1.3.0 GPU depends on CUDA v8 and cuDNN v6 see https://www.tensorflow.org/install/install_windows for detailed instructions on installing those pre-requisites. If you can't satisfy those requirements, use the cpu version.

v0.3.0 for UE4.17

18 Aug 04:37
Compare
Choose a tag to compare

Changes

Note on GPU release
Tensorflow v1.3.0 GPU depends on CUDA v8 and cuDNN v6 see https://www.tensorflow.org/install/install_windows for detailed instructions on installing those pre-requisites.

v0.2.1 for UE4.16

07 Jun 00:31
Compare
Choose a tag to compare

Changes
-Compile fixes for 4.16. Specifically build.cs changes and raw char* path fixes

v0.2.0 for UE4.15

04 Jun 04:11
Compare
Choose a tag to compare

Changes
-TensorflowComponent input api changed from SendTensorInput to SendJsonInput
-Added CallCustomFunction to TensorflowComponent. This allows you to synchronously call any function in your python module with string args and expecting string return data. Json encoding recommended for native data conversion
-Image struct cleaned up and now includes size information
-Socket.IO updated to 0.6.1, including changes allowing partial struct fill from json objects
-Misc verbosity changes

Note on releases
If you have hardware supporting cuda v8 and cudNN v5 use the gpu version for orders of magnitude faster execution.

If you do not have the appropriate gpu hardware, use the cpu version. Only difference is which tensorflow module is downloaded via pip. This can be changed at any time by changing the dependencies in upymodule.json.

See: https://www.tensorflow.org/install/install_windows for details.