This includes an original implementation ofour paper: Puzzlefusion: Unleashing the Power of Diffusion Models for Spatial Puzzle Solving, NeurIPS (Spotlight) 2023.

Our implementation builds upon the publicly available guided-diffusion repository by OpenAI. To begin, first clone our repository, PuzzleFusion, to your local machine. Following this, you can install the necessary dependencies by executing the commands provided below

pip install -r requirements.txt
pip install -e .

The Cross Cut dataset can be accessed via this crosscut-data, we utilized the original code from the Crossing cut puzzle Paper code to generate the data. After downloading the data, please place it within a folder named 'datasets'.

For training you can run:

cd scripts
bash script.sh

For testing you can run:

cd scripts
bash script_test.sh

We also have provided checkpoint for easier testing here, you can download that and move it to ./scripts/ckpts/preds

The Voronoi dataset can be accessed via this Voronoi-data, we utilized the om the Vornoi puzzle Generator to generate the data. After downloading the data, please place it within a folder named 'datasets'.

Samples will be saved in ./scripts/outputs and model checkpoints will saved in to ./scripts/ckpts. Scripts for Vornoi dataset can be found in ./scripts/voronoi_scripts. You can move the file inside there to main ./scripts file. Similarry Although codes are almost identical we provided puzzle fusion voronoi version code under the fulder puzzle_fusion/puzzle_fusion_voronoi you can replaze the files there with files inside puzzle_fusion. Voronoi data reader is also there.

After moving files. For training you can run:

cd scripts
bash script.sh

For testing you can run:

cd scripts
bash script_test.sh

We also have provided checkpoint for easier testing here, you can download that and move it to ./scripts/ckpts/preds

You can download MagicPlan dataset from here, dataset follows same license as code.

@inproceedings{
hosseini2023puzzlefusion,
title={Puzzlefusion: Unleashing the Power of Diffusion Models for Spatial Puzzle Solving},
author={Sepidehsadat Hosseini and Mohammad Amin Shabani and Saghar Irandoust and Yasutaka Furukawa},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=Z764QxwETf}
}