DirectML Object Detection Samples

Overview

DirectML provides a high-performance, hardware-accelerated machine learning inference engine for Windows. This section showcases practical samples that demonstrate how to leverage DirectML for various object detection tasks. From popular models like YOLO to custom implementations, these samples are designed to help you accelerate your AI development on Windows.

Key Concepts

• Inference Acceleration: Utilize your GPU for faster object detection predictions.
• Model Compatibility: Run pre-trained models or deploy your own custom architectures.
• Windows Integration: Seamlessly integrate DirectML into your C++, C#, or Python applications.

Featured Samples

Getting Started

To run these samples, you'll need:

• A Windows machine with a compatible GPU.
• The latest Windows SDK installed.
• The appropriate development environment (Visual Studio, etc.).

Each sample provides detailed instructions on setup, build, and execution. We recommend starting with the YOLOv4 sample for a comprehensive understanding.

Code Snippet Example (Conceptual)

Here's a simplified look at how DirectML might be used for inference:

// Initialize DirectML Device and Operator Graph
IDMLDevice* dmlDevice;
// ... create DML device ...

IDMLOperatorInitializer* initializer;
// ... create initializer ...

IDMLOperator* tensorObjectDetectionOp;
// ... create tensor object detection operator ...

// Prepare input tensors
DML_BUFFER_BINDING inputBinding = { inputBuffer, inputBuffer->GetSize() };
// ... bind input data ...

// Prepare output tensors
DML_BUFFER_BINDING outputBinding = { outputBuffer, outputBuffer->GetSize() };
// ... bind output buffer ...

// Create an execution list
IDMLCommandRecorder* commandRecorder;
// ... create command recorder ...

IDMLCommandList* commandList;
// ... create command list ...

// Record the dispatch command
commandRecorder->RecordDispatch(commandList, tensorObjectDetectionOp, &inputBinding, &outputBinding, nullptr);

// Close and execute the command list
// ... execute ...

// Process the results from the output buffer