Skip to content

OpenVINO Runtime Engine for swatahVision

This code creates a custom runtime engine that allows OpenVINO models to run inside the swatahVision framework.

In simple terms, this code helps the system:

  1. Load an OpenVINO model
  2. Prepare an image so the model can understand it
  3. Run the model on the image
  4. Return the prediction results

What is OpenVINO?

OpenVINO is a toolkit developed by Intel that helps run AI models faster and more efficiently, especially on:

  • CPUs
  • GPUs
  • Intel hardware

It is commonly used for computer vision tasks such as:

  • Image classification
  • Object detection
  • Face recognition

Purpose of This Code

This file creates a class called:

OpenVinoRuntimeEngine

This class connects:

swatahVision → OpenVINO Runtime

This allows swatahVision to run OpenVINO models easily.

Libraries Used

This code uses the following Python libraries:

  • openvino → runs the OpenVINO model
  • numpy → handles numerical data
  • opencv (cv2) → processes images

Class: OpenVinoRuntimeEngine

This class extends the base class:

RuntimeEngine

It defines how the system should:

  • load a model
  • prepare the input image
  • run the model
  • return the results

1. Loading the Model

Function:

load()

This function loads the OpenVINO model into memory.

Steps performed:

  1. Select the hardware (CPU or GPU)
  2. Create an OpenVINO Core object
  3. Read the model file
  4. Compile the model for the selected device

Example hardware options:

CPU → runs on processor
GPU → runs on graphics card

After loading the model, the function also collects information about:

  • input names
  • input shapes
  • input data types
  • output names

2. Running the Model (Inference)

Function:

infer()

This function performs the actual prediction.

Steps:

  1. Receive an input image
  2. Prepare the image using preprocessing
  3. Send the image to the compiled model
  4. Get the prediction results

The function returns:

raw_output
meta_information

3. Getting Model Information

Function:

get_model_info()

This function extracts important details from the model.

It collects:

  • Input names
  • Input shapes
  • Input data types
  • Output names

This information helps the engine understand what format the model expects.

4. Image Preprocessing

Function:

preprocess()

Before sending an image to the model, the image must be prepared correctly.

This function performs several steps to prepare the image.

Step 1: Resize the Image

The image is resized to match the size expected by the model.

Step 2: Keep Image Proportions (Letterbox)

Instead of stretching the image, the code keeps the original shape.

This is done using a technique called letterboxing.

The process:

  1. Resize the image
  2. Add padding around the image

This prevents distortion.

Step 3: Rearrange Image Format

Normally images are stored as:

(H, W, C)
Height, Width, Channels

But most AI models expect:

(C, H, W)
Channels, Height, Width

So the code rearranges the image format.

Step 4: Handle Single or Multiple Images

The engine can process:

  • a single image
  • multiple images at once

If needed, the code automatically adds a batch dimension.

Step 5: Convert Image Data Type

Different models expect different input types such as:

  • float32
  • uint8

The code automatically converts the image to the correct type.

Letterbox Function

The letterbox() function resizes an image without stretching it.

Steps performed:

  1. Calculate how much the image should scale
  2. Resize the image
  3. Add padding to match the required size

Example:

Original image:

400 × 300

Model input size:

640 × 640

The code:

  • resizes the image
  • adds padding around it

This keeps the image looking natural.

Metadata Returned

The preprocessing function also returns metadata information.

This includes:

scale
padding_x
padding_y

This data can later be used to adjust predictions back to the original image size.

Overall Workflow

The full process works like this:

Load Model
      ↓
Receive Image
      ↓
Preprocess Image
      ↓
Run Model
      ↓
Return Prediction

Why This Code is Useful

This engine makes it easier to run OpenVINO models in swatahVision.

It provides:

  • automatic image preprocessing
  • support for CPU and GPU
  • support for batch inputs
  • simple model execution

Summary

This code acts as a bridge between:

swatahVision Framework
        ↓
OpenVINO Runtime Engine
        ↓
AI Model

It simplifies the process of:

  • loading models
  • preparing images
  • running predictions
  • retrieving results

This makes it easier to build computer vision applications using OpenVINO models.