Hackatum2022 Huawei Challenge

OHero

Table of Contents

1. Android Native Mobile App
2. Django Server
3. Pytorch CNN Pipeline

Android Native Mobile App

The Android app is built using the Kotlin language. We benefited from the Google's AR tutorials and source code (https://developers.google.com/ar/develop/java/session-config). Icons used in our repo is take from the svgrepo.com.

Django Server

Start server to handle image post requests and run CNN model to detect which class the environment belong. After runnning below commands HTTP requests should be done to base root endpoint of the server e.g 127.0.0.1:8000. Requests should contain raw byte image with Base64 encoding.

cd server
pip3 install django
python3 manage.py runserver

Testing Object Classification pipeline

Mobilenet_v3_large Transfer Learning

Training and Validation

After running below provide parent directory for the corresponding train/val dataset.

cd server
pip3 install -r requirements.txt
python3 -m ml.training


// e.g : path/to/parent/val
//   path/to/parent/train
//   CLI input should /path/to/parent

---

Inference

After running below provide parent directory for the corresponding issues dataset.

cd server
python3 -m ml.testing


// e.g : path/to/parent/issues.csv
//   path/to/parent/*.jpg
//   CLI input should be /path/to/parent

---