CS449-NeurASP

Course project for CS449

• To demonstrate that reasoning can help to identify perception mistakes by neural networks that violate semantic constraints, which in turn can make perception more robust.

Team

• Aniket Pokle (20D070011)
• Vedang Gupta (200100166)

Getting Started

• Setting up the environment

conda create -n myenv
pip3 install -r requirements.txt
conda activate myenv

• To train the digit recognition model

cd digit_detector
python3 model.py -train

The model is saved at assets/model.pth

• Solve Sudoku and it's Varients

python3 sudoku.py <image_path>

Generating Unique Sudoku

Solves the sudoku in the image and and recursively removes digits from the solved sudoku till a non-unique sudoku is obtained

available options are:

1. naive
2. lrtb

python3 generate_sudoku.py option

Available varients of Sudoku

• Anti-Knight Sudoku : No number repeats at a knight move
• Sudoku-X : No number repeats at the diagonals

Reference

• https://www.ijcai.org/proceedings/2020/0243.pdf