Content Table:

• Project Description
• Main Features
• Quick Start
  • Installation
  • Using the Models
• User's guides
• Authors
• License

Project Description

VIAsegura is a library that helps to use artificial intelligence models developed by the Inter-American Development Bank to automatically tag items on the streets. The tags it places are some of those needed to implement the iRAP road safety methodology.

These models require images with the specifications of the iRAP projects. This means that they have been taken every 20 meters along the entire path to be analyzed. In addition, some of the models require images to be taken from the front and others from the side of the car. The models yield 1 result for each model for groups of 5 images or fewer.

So far, 15 models compatible with the iRAP labeling specifications have been developed and are specified in the table below.

Some of the models can identify all the classes or categories, others can help you sort through the available options.

Main Features

Some of the features now available are as follows:

• Scoring using the models already developed
• Grouping by groups of 5 images from an image list
• Download models directly into the root of the package

Quick Start

Environment Setup

It is advisable to install the library in a Python virtual environment. If you use conda, you can do so with the following command:

conda create -n viasegura python=3.10
conda activate viasegura

Supported versions of Python are 3.9, 3.10, 3.11 and 3.12. For more information on how to install conda, you can visit the official documentation

Installation

To install the package you can use the following commands in the terminal

pip install viasegura

To download the models use this link from the repostitory. Put the downloaded file on a models folder and decompress it with the following command:

tar -xzvf models.tar.gz

The path where the models are must be models/models_artifacts Remember to put that path every time you instantiate a model so that you can find the artifacts you need to run them.

Using the Models

In order to make the instance of a model you can use the following commands

from viasegura import ModelLabeler

frontal_labeler = ModelLabeler('frontal')  # or 'lateral' 

You can use either "frontal" or "lateral" tag in order to use the group of models desired (see table above)

Also, you can specify which models to load using the parameter model filter and the name of the models to use, (see the table above):

from viasegura import ModelLabeler

frontal_labeler = ModelLabeler('frontal', model_filter=['delineation', 'street_lighting', 'carriageway']) 

In addition, you can make it work using the GPU specifying the device where the models are going to run, for example

from viasegura import ModelLabeler

frontal_labeler = ModelLabeler('frontal', device='/device:GPU:0') 

You can modify the devices used according to the TensorFlow documentation regarding GPU usage.

For a full example of use on this package, you can see this notebook. In the notebooks folder you can also find example images to test the execution of the models.

Users Guide

I invite you to visit the manual to understand the scope of the project and how to make a project from scratch using the viasegura models.

Authors

This package has been developed by:

Jose Maria Marquez Blanco
Joan Alberto Cerretani
Victor Durand

License

The distribution of this software is according to the following license