Skip to content

Latest commit

 

History

History
116 lines (78 loc) · 5.09 KB

README.md

File metadata and controls

116 lines (78 loc) · 5.09 KB

Intel® OpenVINOTM integration with TensorFlow Python Object Detection Example

This example demonstrates how to use Intel® OpenVINOTM integration with Tensorflow to recognize and detect objects in images and videos.

Demo showcased in this example

Object detection using YOLOv3 model converted from Darknet to detect objects in a given image, video, directory or camera input.

Setup for the examples

Before you proceed to run the examples, you will have to clone the openvino_tensorflow repository to your local machine. For this, run the following commands:

$ git clone https://github.com/openvinotoolkit/openvino_tensorflow.git
$ cd openvino_tensorflow
$ git submodule init
$ git submodule update --recursive

Python implementation for object detection

For this example, we assume that you've already:

  • Installed TensorFlow on your system
  • Installed Intel® OpenVINOTM integration with Tensorflow on your system

Refer to this page for quick installation using pip.

Install the pre-requisites

$ cd <path-to-openvino_tensorflow-repository>/examples
$ pip3 install -r requirements.txt

The TensorFlow Yolo v3 model used in this demo is not packaged in the repository because of its size. So, follow the instructions below to convert the model from DarkNet to TensorFlow and download the labels and weights to the <path-to-openvino_tensorflow-repository>/examples/data directory in your cloned repo of openvino_tensorflow:

Please note: The instructions below should not be executed in an active virtual environment. The convert_yolov3.sh script activates a python virtual environment for conversion.

$ cd <path-to-openvino_tensorflow-repository>/examples/TF_1_x
$ chmod +x convert_yolov3.sh
$ ./convert_yolov3.sh

Once completed, the <path-to-openvino_tensorflow-repository>/examples/data folder will contain following files needed to run the object detection example:

  • coco.names
  • yolo_v3_darknet_1.pb

Run the object detection example using the instructions below:

$ cd <path-to-openvino_tensorflow-repository>
$ python3 examples/TF_1_x/object_detection_sample.py

Note: use --no_show flag to disable the application display window. By default the display window is enabled.

This uses the default example image that comes with this repository, and should output something similar as below and the result is written to detections.jpg:

In this case, we're using the default image of Admiral Grace Hopper. As you can see, the network detects and draws the bounding box around the person correctly.

Next, try it out on your own image by passing the path to your new image. You can provide either absolute or relative path e.g.

$ python3 examples/TF_1_x/object_detection_sample.py --image=<absolute-or-relative-path-to-your-image>

If you add the new image (e.g, my_image.png) to the existing <path-to-openvino_tensorflow-repository>/examples/data directory in the openvino_tensorflow repository, it will look like this:

$ python3 examples/TF_1_x/object_detection_sample.py --image=examples/data/my_image.png

To see more options for various backends (Intel® hardware), invoke:

$ python3 examples/TF_1_x/object_detection_sample.py --help

Next, try it out on your own video file by passing the path to your input video. You can provide either absolute or relative path e.g.

$ python3 examples/TF_1_x/object_detection_sample.py --input=<absolute-or-relative-path-to-your-video-file>

If you add the new video (e.g, examples/data/people-detection.mp4) to the existing <path-to-openvino_tensorflow-repository>/examples/data directory in the openvino_tensorflow repository, it will look like this:

$ python3 examples/TF_1_x/object_detection_sample.py --input=examples/data/people-detection.mp4

For using camera as input use --input=0. Here '0' refers to the camera present at /dev/video0. If the camera is connected to a different port, change it appropriately.

Note: The results with input as an image or a directory of images, are written to output images. For video or camera input use the application display window for the results.

To try on the yolo_v3_160 model for faster inference follow the below steps

Please note: The instructions below should not be executed in an active virtual environment. The convert_yolov3_160.sh script activates a python virtual environment for conversion.

$ cd <path-to-openvino_tensorflow-repository>/examples/TF_1_x
$ chmod +x convert_yolov3_160.sh
$ ./convert_yolov3_160.sh

Run the object detection example using the instructions below:

$ cd <path-to-openvino_tensorflow-repository>
$ python3 examples/TF_1_x/object_detection_sample.py --input_height 160 --input_width 160 --graph "examples/data/yolo_v3_darknet_160.pb" --input_layer "inputs" --output_layer "output_boxes" --labels "examples/data/coco.names"