Hello Reshape SSD C++ Sample

This sample demonstrates how to execute an inference of object detection networks like SSD-VGG using Synchronous Inference Request API, input reshape feature and implementation of custom extension library for CPU device (CustomReLU kernel).

Hello Reshape SSD C++ sample application demonstrates how to use the following Inference Engine C++ API in applications:

Feature

API

Description

Network Operations

InferenceEngine::CNNNetwork::getBatchSize , InferenceEngine::CNNNetwork::getFunction

Managing of network, operate with its batch size.

Input Reshape

InferenceEngine::CNNNetwork::getInputShapes , InferenceEngine::CNNNetwork::reshape

Resize network to match image sizes and given batch

nGraph Functions

ngraph::Function::get_ops , ngraph::Node::get_friendly_name , ngraph::Node::get_type_info

Go thru network nGraph

Custom Extension Kernels

InferenceEngine::Core::AddExtension

Load extension library

CustomReLU kernel

InferenceEngine::ILayerExecImpl

Implementation of custom extension library

Basic Inference Engine API is covered by Hello Classification C++ sample.

Options

Values

Validated Models

person-detection-retail-0013

Model Format

Inference Engine Intermediate Representation (.xml + *.bin), ONNX (.onnx)

Validated images

The sample uses OpenCV* to read input image (*.bmp, *.png)

Supported devices

All

Other language realization

C , Python

How It Works

Upon the start-up the sample application reads command line parameters, loads specified network and image to the Inference Engine plugin. Then, the sample creates an synchronous inference request object. When inference is done, the application creates output image and output data to the standard output stream.

You can see the explicit description of each sample step at Integration Steps section of “Integrate the Inference Engine with Your Application” guide.

Building

To build the sample, please use instructions available at Build the Sample Applications section in Inference Engine Samples guide.

Running

To run the sample, you need specify a model and image:

NOTES :

  • By default, Inference Engine samples and demos expect input with BGR channels order. If you trained your model to work with RGB order, you need to manually rearrange the default channels order in the sample or demo application or reconvert your model using the Model Optimizer tool with --reverse_input_channels argument specified. For more information about the argument, refer to When to Reverse Input Channels section of Converting a Model Using General Conversion Parameters.

  • Before running the sample with a trained model, make sure the model is converted to the Inference Engine format (*.xml + *.bin) using the Model Optimizer tool.

  • The sample accepts models in ONNX format (*.onnx) that do not require preprocessing.

You can use the following command to do inference on CPU of an image using a trained SSD network:

<path_to_sample>/hello_reshape_ssd <path_to_model> <path_to_image> <device> <batch>

Example

  1. Download a pre-trained model using Model Downloader :

    python <path_to_omz_tools>/downloader.py --name person-detection-retail-0013
  2. person-detection-retail-0013 model does not need to be converted, because it is already in necessary format, so you can skip this step. If you want to use a other model that is not in the Inference Engine IR or ONNX format, you can convert it using the model converter script:

python <path_to_omz_tools>/converter.py --name <model_name>
  1. Perform inference of person_detection.png using person-detection-retail-0013 model on a GPU, for example:

<path_to_sample>/hello_reshape_ssd <path_to_model>/person-detection-retail-0013.xml <path_to_image>/person_detection.png GPU 1

Sample Output

The application renders an image with detected objects enclosed in rectangles. It outputs the list of classes of the detected objects along with the respective confidence values and the coordinates of the rectangles to the standard output stream.

Resizing network to the image size = [960x1699] with batch = 1
Resulting input shape = [1,3,960,1699]
Resulting output shape = [1,1,200,7]
[0,1] element, prob = 0.722292, bbox = (852.382,187.756)-(983.352,520.733), batch id = 0
The resulting image was saved in the file: hello_reshape_ssd_output.jpg

This sample is an API example, for any performance measurements please use the dedicated benchmark_app tool