Place Recognition Python* Demo


This demo demonstrates how to run Place Recognition models using OpenVINO.


Only batch size of 1 is supported.

How It Works

The demo application expects a place recognition model in the Intermediate Representation (IR) format.

As input, the demo application takes:

  • a path to an image

  • a path to a folder with images

  • a path to a video file or a device node of a webcam

The demo workflow is the following:

  1. The demo application reads input frames.

  2. Extracted input frame is passed to artificial neural network that computes embedding vector.

  3. Then the demo application searches computed embedding in gallery of images in order to determine which image in the gallery is the most similar to what one can see on frame.

  4. The app visualizes results of it work as graphical window where following objects are shown.

    • Input frame.

    • Top-10 most similar images from the gallery.

    • Performance characteristics.


By default, Open Model Zoo 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 demo application or reconvert your model using the Model Optimizer tool with the --reverse_input_channels argument specified. For more information about the argument, refer to When to Reverse Input Channels section of Embedding Preprocessing Computation.

Preparing to Run

For demo input image or video files, refer to the section Media Files Available for Demos in the Open Model Zoo Demos Overview. The list of models supported by the demo is in <omz_dir>/demos/place_recognition_demo/python/models.lst file. This file can be used as a parameter for Model Downloader and Converter to download and, if necessary, convert models to OpenVINO IR format (*.xml + *.bin).

An example of using the Model Downloader:

omz_downloader --list models.lst

An example of using the Model Converter:

omz_converter --list models.lst

Supported Models

  • netvlad-tf


Refer to the tables Intel’s Pre-Trained Models Device Support and Public Pre-Trained Models Device Support for the details on models inference support at different devices.


Run the application with the -h option to see the following usage message:

usage: [-h] -m MODEL -i INPUT -gf GALLERY_FOLDER
                                 [--gallery_size GALLERY_SIZE] [--loop]
                                 [-o OUTPUT] [-limit OUTPUT_LIMIT] [-d DEVICE]
                                 [--no_show] [-u UTILIZATION_MONITORS]

  -h, --help            Show this help message and exit.
  -m MODEL, --model MODEL
                        Required. Path to an .xml file with a trained model.
  -i INPUT, --input INPUT
                        Required. An input to process. The input must be a
                        single image, a folder of images, video file or camera
  -gf GALLERY_FOLDER, --gallery_folder GALLERY_FOLDER
                        Required. Path to a folder with images in the gallery.
  --gallery_size GALLERY_SIZE
                        Optional. Number of images from the gallery used for
  --loop                Optional. Enable reading the input in a loop.
  -o OUTPUT, --output OUTPUT
                        Optional. Name of the output file(s) to save. Frames of odd width or height can be truncated. See
  -limit OUTPUT_LIMIT, --output_limit OUTPUT_LIMIT
                        Optional. Number of frames to store in output. If 0
                        is set, all frames are stored.
  -d DEVICE, --device DEVICE
                        Optional. Specify the target device to infer on: CPU or
                        GPU. The demo will look for a
                        suitable plugin for device specified (by default, it
                        is CPU).
  --no_show             Optional. Do not visualize inference results.
                        Optional. List of monitors to show initially.

Running the application with an empty list of options yields the short version of the usage message and an error message.

To run the demo, please provide paths to the model in the IR format, to directory with gallery images, and to an input video, image, or folder with images:

python \
  -m <path_to_model>/netvlad-tf.xml \
  -i <path_to_file>/image.jpg \
  -gf <path>/gallery_folder

> NOTE : If you provide a single image as an input, the demo processes and renders it quickly, then exits. To continuously visualize inference results on the screen, apply the loop option, which enforces processing a single image in a loop.

You can save processed results to a Motion JPEG AVI file or separate JPEG or PNG files using the -o option:

  • To save processed results in an AVI file, specify the name of the output file with avi extension, for example: -o output.avi.

  • To save processed results as images, specify the template name of the output image file with jpg or png extension, for example: -o output_%03d.jpg. The actual file names are constructed from the template at runtime by replacing regular expression %03d with the frame number, resulting in the following: output_000.jpg, output_001.jpg, and so on. To avoid disk space overrun in case of continuous input stream, like camera, you can limit the amount of data stored in the output file(s) with the limit option. The default value is 1000. To change it, you can apply the -limit N option, where N is the number of frames to store.

> NOTE : Windows* systems may not have the Motion JPEG codec installed by default. If this is the case, you can download OpenCV FFMPEG back end using the PowerShell script provided with the OpenVINO install package and located at <INSTALL_DIR>/opencv/ffmpeg-download.ps1. The script should be run with administrative privileges if OpenVINO is installed in a system protected folder (this is a typical case). Alternatively, you can save results as images.

Demo Output

The application uses OpenCV to display gallery searching result. The demo reports

  • FPS : average rate of video frame processing (frames per second).

  • Latency : average time required to process one frame (from reading the frame to displaying the results). You can use both of these metrics to measure application-level performance.