Squad: Virtual Object Detector

This project aims to build object detection for the metaverse by training models with the ImageAI library on web images captured using Puppeteer.

Contents

• 👪 Squad
• 🏗 Initial Setup
• 🏛 License
• Information

👪 Squad

Lead: alextitonis (GitHub)

Grant Proposal: https://forum.algovera.ai/t/virtual-object-detector/25

Notion: https://algovera.notion.site/Squads-194768658a044302a0cdc24d5d758b9d?p=13587733017d4524832fd51f3780969a

🏗 Initial Setup

Set up environment

Open a new terminal and:

#clone repo
git clone https://github.com/AlgoveraAI/squad-virtual-object-detector.git
cd squad-virtual-object-detector

#create a virtual environment
python3 -m venv venv

#activate env
source venv/bin/activate

#Install libraries.
pip install -r requirements.txt

#Edit the variables
Rename the .env.default to .env and edit the variables inside

To run the trainer use the trainer.py For the detector is the detector.py For the editor run the /editor/editor.py file The editor includes functions for train and detection (images, videos) The capturers folder includes varius scripts that can be used to capture images:

• browser.py -> is the wrapper for puppeteer (using pyppeteer) it includes some example functions to enter a website and capture images
• fullScreen.py -> includes a function to capture a screenshot in full screen
• gameWindow.py -> includes a function to capture a screenshot from a window (not only from games) using the window title - windows only
• videoCapturer.py -> includes a function that captures frames from a video and turns them into images

Information

How to train a new model

Create a folder, inside create 2 new folders called test and train, inside each add the same folders with the names of the images (car, human, etc) For example: trees/train/pine trees/train/oak trees/test/pine trees/test/oak

Train images should be atleast 500, while the test folder should include atleast 200 images to train a proper model.

To train it you can either run the trainer.py script using command line or through the editor tab - trainer, upload a zip file with all the folders/images and it will do the job for you!

🏛 License

The license is MIT. Details