ReACT_GCV

Python tool for analyzing images on the Web using the Google Cloud Vision API.

Clone this repository to your own machine by executing the following command in your terminal: git clone https://github.com/rubenros1795/ReACT_GCV.git or download and unzip the repo (top-right corner).

Installation Guide

The Notebooks found in the /notebooks folder require a Pyton installation. In this step-by-step installation guide we explain how to install Anaconda, a Python distribution, and create a virtual environment.

Installing

1. Download Anaconda here: https://repo.anaconda.com/archive/ or through here: https://docs.anaconda.com/anaconda/install/

2. Install Anaconda, for detailed guides see: https://docs.anaconda.com/anaconda/install/

Creating a Virtual Environment

Anaconda installs Python on the system (it is a Python distribution). To run scripts and notebooks it is advised to create a so-called "virtual environment" (VE) for your project. Below we show how to create a VE.

3. Go to the terminal. On Windows you will find it as cmd (search in the bottom-right corner), on Mac and Linux simply as "terminal" in the application overview.

4. Check if you succesfully installed Anaconda by typing conda -v and pressing Enter to execute the terminal command.

5. Create an environment by executing conda create -n yourenvname. This creates a new VE in the anaconda3/envs/ folder.

6. Activate the environment by executing conda activate yourenvname.

Installing Libraries

7. In order to use the notebooks in this repository, we need several seconday Python libraries. These are found in the requirements.txt file. To install the libraries, navigate to the ReACT_GCV folder in the terminal (N.B.: activate the VE first!). Once you are in the right folder, execute conda install --file requirements.txt.

8. Here the trouble starts. Some libaries are not available in the Anaconda channel (the list of libraries that Anaconda knows). To install the missing libaries, first install pip, a Python "librarian".

• Windows: see the instructions here: https://www.liquidweb.com/kb/install-pip-windows/.
• Linux: sudo apt install python-pip.
• Mac: sudo easy_install pip.

9. Install the missing libraries by executing pip install libraryname (for example: pip install purl).

Opening Jupyter Notebooks

10. Open Jupyter Notebooks by simply executing jupyter notebook. Navigate to the /notebooks folder and open the .ipynb file.

Code Instructions

For using the pipeline outside of the Juypter Notebooks, we refer to the scripts in /code The scripts are grouped in three folders: scraping, parsing and analysis.

Scraping

To scrape, edit the folder paths in scrape.py, add the API key and run the script. Setting up a Google Cloud account is explained in the first notebook. The scrape.py script will reupload the scraped images until no unique images are found. An explanation of this iterative pipeline is also found in the notebooks. We found that the API often finds combined images, meaning that the input image is combined with another image. Because this leads to the identification of this second, unwanted, image in the next iteration it is advised to manually check the relevance of the identified image after every iteration. For this reason scrape-manual.py offers the possibility to execute the scraping script for a specific iteration. Edit the folder paths before running the script in the terminal:

python scrape-manual.py --photo example_photo --iteration 1

Parsing

Parsing the data consists of 1) scraping the webpages associated with the identified URLs, 2) extracting texts from the webpages, 3) generating metadata and 4) enriching the texts using Named Entity Recognition. The functions are wrapped in functions.py and can be called over a photo in parse.py.

Analysis

Notebooks and helper functions for various types of analysis are found in /analysis.