TensorFlow 2 Object Detection API on ubuntu 18.04

Set up and running TensorFlow API

---

A "short" tutorial that started off as self notes on how to set-up and running with the TensorFlow Object Detection API. In July 10, 2020 TensorFlow introduced That the Object Detection API officially supports TensorFlow 2. Therefore, an updated version of the tutorial was created to cover itS.

To read the tutorial, visit http://tensorflow-object-detection-api-tutorial.readthedocs.io.

Install TensorFlow 2 on Ubuntu 18.04 OS

---

step 1. Creating a virtual environment

Create a virtual environment in order to install TensorFlow into it without compromising with other projects. I am utilizing Anaconda usually. To create a new virtual environment called tf2.3, for instance. Run the following command to create and activate the environment:

conda create --name tf2.3 python=3.8

conda activate ft2.3

Create a new folder and name it test, for instance. Navigate to this directory $ cd test,

(tf2.3) user@username:~/test$

step 2. (a)-Download

Download or clone TensorFlow repo. from Tensorflow official link in your test directory

git clone https://github.com/tensorflow/models

Then extract all the files from model-master.zip. Then copy research folder from $test/model-master/ to ``

(tf2.3) user@username:~/test/model-master$
(tf2.3) user@username:~/test$scp -r model-master/research . #cp research to test folder
(tf2.3) user@username:~/test$rm -r model-master.zip   # you can delet this folder 
(tf2.3) user@username:~/test$cd research 
(tf2.3) user@username:~/test/research$cp object_detection/colab_tutorials/object_detection_tutorial.ipynb . #copy notebook file to research folder 

Now we have all the required files to start object detection tutorial.

(b)- installing libraries

Start by installing jupyter notebook

pip install jupyter
pip install "tensorflow == 2.3.0" # instlling the latest version
pip install tf_slim

pip install pycocotools
# Successfully installed cycler-0.10.0 cython-0.29.21 kiwisolver-1.3.1 matplotlib-3.3.2 pillow-8.0.1 pycocotools-2.0.2

(c) Protobuf compilation

The TensorFlow object detection API uses protobufs, protocol buffers -- Google's data interchange format (https://github.com/protocolbuffers/protobuf/releases), to configure the models and their training parameters. Before the framework can be used, the protobuf libraries must be compiled, and that requires different steps if you are in a Unix (Linux or Mac) or Windows OS environment.

So for the Linux OS, download protoc-3.13.0-linux-x86_64.zip from protobuf in your test folder and unzip it unzip protoc-3.13.0-linux-x86_64.zip .

# From test/research directory
(tf2.3) user@username:~/test/research$ ~/test/protoc-3.13.0-linux-x86_64/bin/protoc object_detection/protos/*.proto --python_out=.

step 3. running open a jupyter notebook from research path. Accordingly set the relative path in *Loading label map section:

#Note: you are alrady in research folder
PATH_TO_LABELS = 'object_detection/data/mscoco_label_map.pbtxt'  

PATH_TO_TEST_IMAGES_DIR = pathlib.Path('object_detection/test_images') 

if you get error,

import pathlib

The object_detection_tutorial_mod.ipynb will run fine.

---

Credits

Source code (https://github.com/tensorflow/models)

Installation (https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/install.html)

Installation (https://www.tensorflow.org/install)