Materialistic: Selecting Similar Materials in Images

Prafull Sharma · Julien Philip · Michaël Gharbi · William T. Freeman · Frédo Durand · Valentin Deschaintre

SIGGRAPH 2023 (Journal)

This repository contains the code for the paper Materialistic: Selecting Similar Materials in Images, accepted as a journal paper in SIGGRAPH 2023.

Setup

The code is written in python 3.9 and uses PyTorch along with pytorch-lightning for distributed training. Please create a new conda environment with python 3.9 and run the setup.sh script to install all the dependencies.

conda create -n materialistic python=3.9
conda activate materialistic

sh setup.sh

Dataset

Synthetic training dataset can be downloaded here to be stored in data/materialistic_synthetic_dataset.

Real images used for testing can be downloaded here to be stored in data/materialistic_test_dataset.

After running the setup script and placing the dataset and checkpoints in the correct location, the directory structure should look like this:

    .
    |-- code <-- this repository
    |   |-- checkpoints
    |   |-- ALL OTHER CONTENTS OF THIS REPOSITORY
    |
    |-- tensorboard
    |-- data
        |-- materialistic_synthetic_dataset
        |-- materialistic_real_dataset

Pretrained Models

The pretrained models can be downloaded here. Please unzip the file in the checkpoints folder.

Training

python train.py --exp_name materialistic_transformer_model --epochs 30 --batch_size 8 --num_gpus 2  --config ./configs/transformer.cfg --data_dir ../data/materialistic_synthetic_dataset/ --lr 1e-4

Testing

python test.py --exp_name materialistic_checkpoint --method_name materialistic_checkpoint --config ./configs/transformer.cfg --data_dir ../data/materialistic_real_dataset/

Running the demo

python gradio_demo.py

Acknowledgements

Part of the code is borrowed from DPT. We would like to thank Sameer Khurana for thoughtful discussions and feedback.

Citation

@article{10.1145/3592390,
    author = {Sharma, Prafull and Philip, Julien and Gharbi, Micha\"{e}l and Freeman, Bill and Durand, Fredo and Deschaintre, Valentin},
    title = {Materialistic: Selecting Similar Materials in Images},
    year = {2023},
    issue_date = {August 2023},
    publisher = {Association for Computing Machinery},
    address = {New York, NY, USA},
    volume = {42},
    number = {4},
    issn = {0730-0301},
    url = {https://doi.org/10.1145/3592390},
    doi = {10.1145/3592390},
    journal = {ACM Trans. Graph.},
    month = {jul},
    articleno = {154},
    numpages = {14},
    keywords = {segmentation, selection, material}
}

prafull7/materialistic