Skip to content

Latest commit

 

History

History
30 lines (20 loc) · 2.17 KB

README.md

File metadata and controls

30 lines (20 loc) · 2.17 KB
Raw

GROUN - Get Rid of Your Notes

GROUN is an application which can be used to delete the spam related to lecture slides/notes present in our mobile devices. It currently predicts the image to be a lecture slide or not with a validation accuracy of 95 percent.

Architecture

The model is built on the standard LeNet architecture and on a dataset of over 10000 images, further increased through Data Augmentation.

Project Execution

  1. Open the Terminal.
  2. Clone the repository by entering https://github.com/pranay-ar/GROUN.git.
  3. Ensure that Python3 and pip/conda is installed on the system.
  4. Create a virtualenv by executing the following command: virtualenv -p python3 env.
  5. Activate the env virtual environment by executing the follwing command: source env/bin/activate.
  6. Enter the cloned repository directory and execute pip install -r requirements.txt.
  7. To train the model, run the train.py script by using the following command python train.py --dataset <datasetpath> --model weights.model
  8. For visualising the result of the trained model, execute the following command: python test.py --model slides.model \ and it will ask for the image location you want to test the model on.
  9. Enter the image location as follows: --image <location> and the model will display the results.
  10. For running the model on a folder consisting of all kinds of images, then run the batch.py and provide the name of the folder to be cleaned in the notes path and after the script is succesfully run, it will create a seperate folder for lecture slides and migrate all the probable pictures of slides to that folder thereby cleaning the image folder.

Performance Overview

Miscellaneous

Inorder to create a custom dataset, the following link can be used to create a script to download mutliple images at once for a single query.

I am currently working on creating a web-application to automate this process of deleting the spam images present in our devices. Feel free to open an issue to improve an existing feature or to collaborate on this together.