The Black Friday Creator Sale is here! Save up to 70% on selected products now through December 3.

Velox Neuro - Neural Network Inference Engine

Velox XR - Code Plugins - Oct 17, 2023

Utility plugin for running inference using ONNX Runtime native library

  • Supported Platforms
  • Supported Engine Versions
    5.1 - 5.3
  • Download Type
    Engine Plugin
    This product contains a code plugin, complete with pre-built binaries and all its source code that integrates with Unreal Engine, which can be installed to an engine version of your choice then enabled on a per-project basis.

**** UPDATE ****

Velox Neuro plugin version 0.65b Update:

  • Mac Platform Support
  • Android Platform Support
  • General performance / stability update
  • Fixed FillTextureFromTensor / ConvertTensorToTexture to support tensors with more than 1 channel
  • Added SaveTensorAsImage to save tensors as PNG file
  • Added PermuteTensorChannels function
  • Added support for tensors without batching information
  • Added support for single channel textures in ConvertTextureToTensor
  • ONNX Runtime updated to version 1.16.2
  • Default search path "MLModels" in the Project's Content for standalone builds


Velox Neuro is an open-source code plug-in that provides a set of utilities for running inference using ONNX Runtime native library.


This plugin supports ONNX (Open Neural Network Exchange), which is an open-source machine learning model format widely used.


  1. Create a Blueprint
  2. Create an instance of Inference Engine with your chosen ONNX model file
  3. Assign Tensor Inputs from Textures
  4. Execute Inference
  5. Get Output as Textures


Many ML frameworks such as PyTorch and TensorFlow can export its model in ONNX format.


Many trained models are available on ONNX Model Zoo.


Get example ONNX files from here:

  • MiDaS_3_1_Swin2_L.onnx
  • MiDaS_3_1_Swin2_T.onnx
  • rvm_mobilenetv3_fp32.onnx
  • rvm_resnet50_fp32.onnx
  • yolov7.onnx


Download our Demo Project (compatible with 5.1, 5.2, 5.3) with example of loading any media sources and running inference with:

  • Object Detection from a video returning a 2D Region of Interest (RoI)
  • Mask Generation
  • Depth estimation
  • Real-time mesh animation with depth projection

Read this Documentation before using the Demo Project.


Video of using a webcam in the Demo Project with Velox Neuro.


Velox Neuro plugin is used by Velox Player plugin for importing volumetric videos in the VLX format taken by iOS devices with LiDAR.


The plugin supports running inference on Windows using:

  • CPU
  • CUDA
  • DirectML


Visit our Product page.

Technical Details

Features:

  • Loads ONNX models at runtime.
  • Runs ONNX models.
  • Supports hardware acceleration by DirectML on DirectX 12 capable GPUs.
  • Supports hardware acceleration by CUDA on supported NVIDIA GPUs.


Code Modules:

  • OnnxRuntime (Runtime)
  • VeloxNeuro (Runtime)


Number of C++ Classes: 7

Network Replicated: No

Supported Development Platforms: Windows

Supported Target Build Platforms: Windows


Documentation: LINK

Example Project: LINK

Discord server: Link