SDAIA's Smartathon 2023 - Theme 1 - Team Pattrn

This project is a submission for SDAIA's Smartathon 2023, Theme 1. It aims to build a solution to detect and classify objects such as graffiti, faded signage, potholes, garbage, construction road, broken signage, bad streetlight, bad billboard, sand on road, clutter sidewalk, and unkept facades.

Table of Contents

• Main Notebook
• Presentation
• Technical Report
• Perform Inference using Our Model
• Model Weights
• Authors
• Code Contribution

Main Notebook

The main notebook details our complete work and how to replicate the training, and adds more details on the inference. The notebook can be found here.

Presentation

The presentation showcasing an overview of the project and work can be found here.

Technical Report

The technical report detailing the methodology and procedures can be found here.

Perform Inference using Our Model

1. Clone this repository:
   

git clone https://github.com/MeshalAlamr/smartathon-theme-1.git

2. Download the Smartathon Theme 1 dataset and unzip it in the root directory.

3. Install the project requirements:

cd smartathon-theme-1
pip install -r requirements.txt

4. To run the inference on the test.csv images of the contest using our model run the following:

python inference.py --model_name pattrn --segments 8

This will generate the output csv file containing the labels and bounding boxes for the visual pollutions in the results folder.

Note: In case you faced memory issues during inference, try to increase the segments parameter.

Model Weights

The final model weights can be found here.

Authors

• Meshal Alamr

• Abdulmajeed Almuarik

• Saud Alrasheed

• Hussain Alkhayat

• Wael Mobeirek

• Kumail Alhamoud

Code Contribution

• Meshal Alamr

• Notebook, Training, Inference, and Report.

• Abdulmajeed Almuarik

• Data Preparation, Training, and Presentation.

Research Ideas

• Meshal Alamr

• Abdulmajeed Almuarik

• Saud Alrasheed

• Hussain Alkhayat

• Wael Mobeirek

• Kumail Alhamoud