JunctionX Seoul 2020 Hackathon

🥉 SIA Track 3rd winner

SmartMonitor

Detect damaged building by natural disasters and get a brief information about it.

Contributors

• Nikita Rusetskii (me)
• Konstantin Shusterzon
• Kirill Zenin

Technologies used

For development of the web app we used following technologies:

• ASP.NET Core;
• Bulma;
• NewsAPI;

Machine learning side was developed with PyTorch. We used Res-Net50 pretrained on COCO dataset.

How it works

1. Pick up a natural disaster;
2. Check news and photos with detected by ResNet damaged houses

Demo

To watch demo click on the image below

PyTorch

In /PyTorch folder you can see two files:

• predict.py works with trained ResNet, gets two images as input and returns 2 images with detected houses
• torch_object_detection.py - ResNet is build, dataset is collected and network is trained

Presentation

Check on Google Docs

Future plans

Due to the lack of experience we haven't managed to successfully integrate it with Azure and deploy it but we're looking forward to improve our project! Our future plans are:

• Azure integration;
• Adding data storage;
• Better natural disaster news aggregation;
• Retraining neural network to be able to detect other objects and show more detailed information about damage;
• Azure/DigiatOcean Deployment

Credits

This amazing pic in the header is made by MacroVector